Sex Matters: Gender Bias in the Mutual Fund Industry ∗
Alexandra Niessen-Ruenzi and Stefan Ruenzi
July 2015
ABSTRACT
We document significantly lower inflows in female-managed mutual funds than in male-managed
funds. This result is obtained with field data and with data from a laboratory experiment. There are
no gender differences in performance. Thus, rational statistical discrimination is unlikely to explain
the fund flow effect. We conduct an implicit association test and find that subjects with stronger
gender bias according to this test invest significantly less into female-managed funds. Our results
suggest that gender bias affects investment decisions and thus offer a new potential explanation for
the low fraction of women in the mutual fund industry.
JEL-Classification Codes: G23, J71
Keywords: Mutual Funds; Investor Behavior; Gender Bias; Implicit Association Test
∗Both authors are from the University of Mannheim. We have benefited from comments by Vikas Agarwal,Yakov Amihud, Brad Barber, Jonathan Berk, Michael Cavanaugh, John Chalmers, Werner DeBondt, ElroyDimson, John Griffin, Mark Hulbert, Jayant Kale, Shmuel Kandel, Markku Kaustia, Shimon Kogan, CameliaKuhnen, Alok Kumar, Jerry Parwada, Andrei Simonov, Laura Starks, Russ Wermers, participants at the HarvardKennedy School Conference on ’Closing the Gender Gap’, the European Economic Association Meetings, theFMA Meetings, the EFM Symposium on Behavioral Finance, the Helsinki Finance Summit on Investor Behavior,as well as seminar participants at Georgia State University, University of Texas (Austin), University of NewSouth Wales, University of Sydney, Australian National University, Maastricht University, Erasmus Universityof Rotterdam, and University of Copenhagen. An earlier version of this paper was judged the best conferencepaper at the German Finance Association meetings and the annual meetings of the Society for the Advancementof Behavioral Economics. This paper was judged the best paper at the Rothschild Caesarea Conference, 2012.All errors are our own. Please address all correspondence to [email protected], tel. ++49-(0)621-181-1595, University of Mannheim, L9, 1-2, 68131 Mannheim, Germany.
Sex Matters: Gender Bias in the Mutual Fund Industry
July 2015
ABSTRACT
We document significantly lower inflows in female-managed mutual funds than in male-managed
funds. This result is obtained with field data and with data from a laboratory experiment. There are
no gender differences in performance. Thus, rational statistical discrimination is unlikely to explain
the fund flow effect. We conduct an implicit association test and find that subjects with stronger
gender bias according to this test invest significantly less into female-managed funds. Our results
suggest that gender bias affects investment decisions and thus offer a new potential explanation for
the low fraction of women in the mutual fund industry.
JEL-Classification Codes: G23, J71
Keywords: Mutual Funds; Investor Behavior; Gender Bias; Implicit Association Test
1 Introduction
Why are there so few women in the financial industry? The fraction of female fund managers in charge
of a single managed U.S. equity fund has hovered around a very low level of about 10% for the last 20
years. While various reasons like hiring discrimination against women (Goldin and Rouse (2000)), self se-
lection of women into other professions (Polachek (1981)), less competitive environments (Niederle and
Vesterlund (2007) and Sutter and Gatzle-Rutzler (2014)), or career interruptions (Bertrand, Goldin,
and Katz (2010)) can contribute explaining the low fraction of women in this industry, we suggest
customer-based discrimination as an alternative explanation for this phenomenon (Becker (1971)). Our
starting point is the conjecture that some investors might be subject to gender bias, which eventually
leads to lower money flows in female-managed funds.1 Consequently, hiring women as fund managers
would be less attractive for fund companies, as they generate their profits from fees charged on assets
under management. This paper presents results from an empirical study, from an experimental invest-
ment task, and from an implicit association test (IAT) that support the idea that investors are subject
to such a gender bias.
Our empirical investigation using field data from all single-managed U.S. equity mutual funds from
1992 to 2009 shows that female-managed funds experience significantly lower money inflows than male-
managed funds. The growth rates of female-managed funds are about one third lower than those of
male-managed funds. Furthermore, interactions of manager gender with fund performance indicate that
a fund profits less from good past performance in terms of inflows if it is managed by a woman.
There are two main reasons why investors might shy away from female fund managers: (rational)
statistical discrimination (e.g., Phelps (1972)) or (irrational) prejudice against female fund managers
due to gender bias (e.g., Becker (1971)). If female fund managers underperform or show other un-
desirable investment behavior, it would be rational for investors to use the manager’s gender as a
signal of their investment skills; eventually they would statistically discriminate against female fund
managers by investing less in their funds. However, we find no evidence for gender differences among
fund managers that would support the view that shying away from female managers could be rational:
their investment styles are more persistent over time than those of male fund managers, while average
1Anecdotal evidence from interviews with fund managers suggests that this is indeed the case: asked why female-managed funds attract less capital, one fund manager stated: “There’s something that prevents people from being totallycomfortable about signing their money over to a woman...a lot of negatives are applied.” (National Council for Researchon Women (2009)).
1
performance is virtually identical and male fund managers exhibit less performance persistence. Thus,
if anything, fund investors should prefer female fund managers.
In our regressions, we control for differences in past fund performance, fund and fund company
characteristics such as size, and differences in characteristics of the fund manager other than the
manager’s gender. Furthermore, we conduct a matched sample analysis and obtain similar results. To
address the concern that fund companies might assign female managers to funds that are less attractive
to fund investors for reasons that we cannot explicitly control for, we look at manager changes. We find
that fund flows decrease significantly if a male manager is replaced by a female manager, but not if a
female manager is replaced by a male manager. Additional analysis shows that our results also cannot
be explained by a potentially better access of male managers to male-dominated institutional investor
networks, by potential ’macho-ism’ of brokers who steer investors away from female-managed funds, or
by differences in the extent of media coverage.
There are two possible concerns regarding our results. First, one might wonder whether fund in-
vestors are even aware of who is managing their fund. Second, according to the equilibrium arguments
made in Berk and Green (2004), one could argue that female fund managers would underperform if
they would grow larger because of diseconomies of scale and that the low inflows into female-managed
funds could thus be a rational equilibrium outcome. In our later discussion we address these con-
cerns and present supporting evidence for our postulated gender bias channel and against alternative
explanations.
To further investigate gender bias in investment decisions, we conduct a controlled laboratory
experiment similar to Choi, Laibson, and Madrian (2011). Specifically, participants in the experiment
have to decide how to split a certain amount of money between two index funds. We chose index funds
because the ability of a fund manager to outperform the market is irrelevant for this type of fund. In
the experiment, we keep all information about the fund constant, except for the managers’ names based
on which participants can infer their gender. If participants ignore the manager’s name, as they should
in this setting, we should not find any impact of gender on the chosen investment amount. However, we
observe that subjects in our experiment invest significantly less in the same index fund if the manager
name provided indicates a female manager. The effect is mainly driven by male subjects, while female
subject do not seem to be biased.
2
Finally, to test directly whether there is a gender bias in finance, we conduct an implicit association
test (IAT) with the same subjects who participate in the investment task.2 IATs are an established
experimental method regularly employed by social psychologists to uncover prejudice based on associa-
tions. IATs consist of computerized sorting tasks and allow researchers to measure implicit associations
between concepts (e.g., ’Science’ and ’Liberal Arts’) and group affiliation (e.g., ’Male’ vs. ’Female’)
based on reaction times. External validations of IATs show that they are able to reliably capture preju-
dice and predict behavior (e.g., Greenwald, Poehlman, Uhlmann, and Banaji (2009)). We develop a new
IAT to test for a potential gender bias in finance. Results indicate a bias against women in finance for
most of the subjects in our experiment. Linking the results from the IAT back to subjects’ investment
behavior, we find that subjects with high IAT prejudice scores do indeed invest significantly less in
female-managed funds in the experimental investment task, while subjects for which the IAT does not
indicate any gender bias do not invest less in these funds.
While we cannot provide direct evidence that fund companies consider the lower flows that have to
be expected when hiring a female fund manager, the results from our empirical study as well as from
the experimental investment task and the IAT suggest that this would be a plausible reaction and thus
offer a new customer-based explanation of why we see so few women in the fund industry.
We also discuss why we then see any women in this industry at all. We provide evidence consistent
with the notion that some investor groups are not biased against women or have a preference to invest
with funds from companies that employ female fund managers, e.g., due to diversity policies. Our results
show that the male-managed funds of companies that employ at least one female manager experience
higher inflows, i.e., there is a positive spill-over effect of employing female fund managers on the other
funds in the family.
Our study contributes to the large literature on the determinants of mutual fund performance and
inflows. Chevalier and Ellison (1999) and Baks (2003) examine the impact of fund manager charac-
teristics on fund performance (without a focus on gender). Papers on the determinants of fund flows
mainly focus on the impact of past performance (e.g., Sirri and Tufano (1998), among many others).
Atkinson, Baird, and Frye (2003) look at a small sample of bond funds, but generally find—with the
exception of the first year a female manager manages a fund—no impact of gender on flows.
The idea that mutual fund investors are subject to behavioral biases is examined in Bailey, Kumar,
and Ng (2011). Kumar, Niessen-Ruenzi, and Spalt (2015) find a negative impact of foreign sounding
2A short introductory note on the IAT is Carney, Nosek, Greenwald, and Banaji (2007).
3
names on mutual fund flows, consistent with xenophobia driving investor behavior. Our paper is the first
to show that gender bias of investors has an important impact on investment decisions, too. Thereby, it
contributes to the large sociopolitical debate on gender stereotyping (e.g., Neumark (1996), Bertrand
and Hallock (2001), Newton and Simutin (2014)) by showing that gender bias is also an issue in the
financial industry. Gender issues in a financial context are not widely researched, but at least three
important related papers exist: Wolfers (2006) examines stock market returns of firms with female
CEOs and male CEOs and finds them to be identical, Kumar (2010) analyzes financial analysts and
finds that female analysts provide better estimates and that markets react stronger upon their reports,
while Green, Jegadeesh, and Tang (2009) find that the forecast precision of female analysts is lower
than that of male analysts.
Furthermore, we relate to the broad literature on gender differences in general (e.g., Barber and
Odean (2001), Croson and Gneezy (2009), and Adams and Funk (2012)) and to the general literature
on the influence of manager characteristics on economic outcomes (e.g., Betrand and Schoar (2003)).
Our evidence also complements the earlier literature on customer-based discrimination, which mainly
focuses on racial discrimination (e.g., Nardinelli and Simon (1990), and Holzer and Ihlanfeldt (1998)) in
non-financial contexts. To the best of our knowledge, our paper is the first that analyzes customer-based
gender discrimination.
Finally, our paper contributes to the finance literature methodologically by introducing the IAT
method to the field, which has not been used in finance before.3
2 Data and summary statistics
Our primary data sources are the CRSP Survivor-Bias-Free Mutual Fund Database as well as the
Morningstar Direct and Morningstar Principia mutual fund databases. While the former contains high
quality data on fund performance, the latter is more precise with respect to manager identities and
manager information (Massa, Reuter, and Zitzewitz (2010)). The CRSP database covers virtually all
U.S. open-end mutual funds and provides information on fund returns, fund management structures,
total net-assets, investment objectives, fund managers’ identity, and other fund characteristics. The
Morningstar databases provide information on fund managers including their age and education.
3There are only two papers we are aware of that use IATs in the economics literature: Bertrand, Chugh, and Mul-lainathan (2005) use an IAT to examine hiring discrimination against African-Americans and Beaman, Chattopadhyay,Duflo, Pande, and Topalova (2009) apply an IAT to measure attitudes towards female leaders.
4
We focus on actively managed equity funds that invest more than 50% of their assets in stocks and
exclude bond and money market funds. This allows us to focus on a homogenous group of funds for
which we can easily compare performance. We aggregate the SI and Lipper objective codes contained
in the CRSP database to define the market segment in which a fund operates. This leaves us with
twelve different equity fund segments.4 Following Daniel, Grinblatt, Titman, and Wermers (1997), we
aggregate all share classes of the same fund to avoid multiple counting. Baer, Kempf, and Ruenzi (2011)
show that team managed funds and single managed funds behave differently. Thus, we concentrate on
single managed funds and exclude all team managed funds and funds for which Morningstar gives
multiple manager names from our analysis. Our study covers the time period from January 1992—the
year from which on detailed fund information data are available in the CRSP mutual fund database—to
December 2009.
We identify fund managers’ gender based on their first names as given in the Morningstar databases.
Massa, Reuter, and Zitzewitz (2010) show that information on fund managers as reported by Morn-
ingstar much better match the content of official regulatory filings, i.e., it is more reliable than fund
manager information as provided by the CRSP database. Therefore, we use fund manager names from
Morningstar to identify the fund manager’s gender. Overall, we are able to identify the gender of the
fund manager in 99.39% of all cases based on the procedure described in detail in Appendix A. Infor-
mation on the age of a fund manager, whether a fund manager obtained a Bachelor, MBA, or PhD
degree, and whether a fund manager obtained a professional qualification (mainly Chartered Financial
Analyst, CFA, but also others, e.g., Chartered Financial Planner, CFP, or Certified Public Accountant,
CPA) are collected from fund manager biographies in Morningstar Principia and Morningstar Direct,
Capital IQ, and from internet searches. Data on the media coverage of fund managers based on the
number of newspaper articles in which a manager appears are obtained from the LexisNexis database.
A detailed description of all variables used in our later analysis is contained in Appendix B. Appendix
C contains a description of the media coverage data collection process.
Our final sample contains 16,509 fund year observations, out of which 14,804 (89.67%) have a
male manager and 1,705 (10.33%) have a female manager. Figure 1 plots the total number of male
and female-managed funds as well as the fraction of female-managed funds over our sample period. It
4Specifically, we use the following twelve equity fund segments: AG (Aggressive Growth), BAL (Balanced Funds),EM (Emerging Markets), GE (Global Equity), GI (Growth and Income), IE (International Equity), IN (Income), LG(Long-term Growth), RE (Regional Funds), SE (Sector Funds), UT (Utility Funds), and TR (Total Return).
5
shows that the fraction of female-managed funds is low and constant at around 10% over our whole
sample period.
Panel A of Table 1 reports summary statistics for various fund and manager characteristics for the
sample of funds that we later use in our regression analysis. In panel B of Table 1, we report differences
in fund characteristics between female and male-managed funds in our sample for the most important
variables.
The univariate comparison shows that female-managed funds get significantly lower money inflows
than male-managed funds and female managers are responsible for significantly smaller funds, while
the mean age of female-managed funds is slightly higher than the mean age of male-managed funds.
With respect to fees, we find that 12b-1 fees are significantly higher for female-managed funds than for
male-managed funds. We also find that female managers trade significantly less than male managers.
Female manager have a slightly better average performance based on Sharpe Ratios, but there is no
difference in average performance based on factor alphas or raw returns and no significant difference
in average risk. Female fund managers have a significantly lower tenure with a particular fund and
they are significantly less likely than male fund managers to hold a PhD degree. Finally, the media
coverage of female fund managers is significantly lower than that of male fund managers. While male
fund managers are covered more than twice per year on average, female managers are mentioned less
than once per year in the public press.
3 Do investors care about the manager’s gender? - Empirical evi-
dence
3.1 Fund flows and manager gender
We start our empirical analysis by examining whether female-managed funds attract lower inflows than
male-managed funds. We relate net-inflows into a fund, FundF lowsi,t to a female dummy variable,
Femalei,t, that equals one if the manager of fund i in year t is female, and zero otherwise. As control
variables, we add several characteristics that have proven to influence fund flows. Specifically, we control
for the influence of past performance, FundReti,t−1, on fund flows. We also include lagged fund size,
FundSizei,t−1, the fund’s annual turnover ratio, TORatioi,t−1, the fund’s age, FundAgei,t−1, lagged
fund risk, FundRiski,t−1, as well as a fund’s lagged expense ratio in percent, ExpRatioi,t−1, in our
6
regression.5 All variables are defined in more detail in Appendix B. Sialm and Tham (2015) show that
the stock market performance of mutual fund companies can impact the flows of the affiliated funds.
To account for the impact of this effect and other characteristics of the fund company on inflows,
we additionally include percentage flows in the respective fund’s management company c in year t,
CompanyF lowc,t. Factors affecting flows of new money into the whole segment of the fund are consid-
ered by adding the percentage of flows in the respective market segment k in year t, SegmentF lowk,t.6
We estimate our empirical models by applying a pooled regression approach with standard errors clus-
tered at the fund level and time, segment, and fund company fixed effects as well as Fama and MacBeth
(1973) regressions. Estimation results are presented in Table 2.
Our findings show that flows into female-managed funds are significantly lower than those into
male-managed funds. The impact of the female dummy is negative and statistically significant at the
1% level in all model specifications. The effect is also economically meaningful: depending on the model
specification, the estimate for the influence of the female dummy shows that a female-managed fund
grows by about 10% to 16% p.a. less than a comparable fund that is managed by a male fund manager.
Given that the average fund in our sample grows by 28% p.a. (see Table 1), this means that a female-
managed fund grows by 35% to 50% less (in relative terms) than a comparable fund that is managed
by a male fund manager.
In Column 1 we control for the impact of past performance by just including the past return of the
fund,7 while in Column 2 (and all following specifications) we additionally include lagged fund flows,
FundF lowsi,t−1. Sirri and Tufano (1998) show that past performance ranks have a nonlinear impact
on fund flows. Thus, in Column 3 and 4 we follow Barber, Odean, and Zheng (2005) and estimate
a quadratic performance flow relationship based on net return ranks and based on Carhart (1997)
four factor alpha ranks.8 We can confirm the convex performance-flow relationship documented in the
literature. More importantly, the impact of the female dummy remains stable.
To address concerns that the performance of funds from different segments is not easily comparable,
in Column 5 we estimate the same model as in Column 3 but focus on a more homogenous subgroup of
5To control for further differences in investment styles, in unreported tests we also include the loadings of fund returnson the Carhart (1997) style factors as explanatory variables. Our results remain unaffected.
6Company flows and segment flows are computed net of the flows into the fund under consideration.7Fund returns are winsorized at the top 1%. Not winsorizing returns does not change the results.8We use performance ranks because Patel, Zeckhauser, and Hendricks (1991) show that ordinal performance measures
can explain fund flows better than cardinal measures. Ranks are calculated for each year and segment separately and areevenly distributed between 0 and 1.
7
funds that exclusively invest in U.S. equities and belong to the segments ’Aggressive Growth’, ’Long-
term Growth’, ’Income’, ’Sector’, and ’Growth & Income’. Results are very similar.
In Column 6 we conduct Fama-MacBeth (1973) regressions using ranks based on returns and in
Columns 7 and 8 we again repeat the standard regression from Column 3 but cluster standard errors
by year or by fund and year, respectively. The impact of the female dummy remains highly significant
and is of similar magnitude across specifications, indicating lower inflows of female-managed funds in
the range of 10% to 11% p.a.
Results in this table show that fund size is one of the main drivers of funds inflows. Although we
control for the linear impact of fund size in all our regressions, the difference in size of female and
male-managed funds (see Table 1) in combination with a possibly non-linear influence of fund size on
fund flows might affect our result. Therefore, in Column 9, we include fund size to the power of two
and three as additional explanatory variables. Our findings are not materially affected.9
Finally, in Columns 10 and 11, we interact our female manager dummy variable with lagged fund
returns as well as with lagged performance ranks and squared lagged performance ranks, respectively.
In Column 10, we find a significantly negative impact of the interaction term, suggesting that flows to
female-managed funds are generally less performance sensitive. In Column 11, the interaction term of
the female dummy with squared past performance is highly significant and positive, while the interaction
term with linear past performance ranks is significantly negative, but smaller in absolute size. This
indicates that a fund profits less from good past performance if it is managed by a woman, while the
punishment for bad performance does not differ much between female and male managers. Irrespective
of the inclusion of the performance interaction terms, the female dummy itself is still significantly
negative and of similar magnitude as before, showing a strong negative level effect.
Regarding our results on the influence of the control variables, they are very uniform across spec-
ifications and generally confirm findings reported in the literature. Overall, our results so far support
the notion that investors exhibit gender bias and prefer male-managed funds.
3.2 Alternative explanations
We now refine our analysis and try to empirically disentangle alternative explanations for the docu-
mented lower inflows into female-managed funds. Results are presented in Table 3.
9We also control for the impact of size by including dummies based on size deciles. Our main result (not reported) isnot affected.
8
First, it is possible that investors prefer certain funds for reasons we do not control for and that
women are more likely to manage such funds – either because they self-select to manage those funds
or because they are assigned to these funds by the fund company. To separate the impact of such fund
characteristics from the impact of gender on fund flows, in Column 1 we look at the impact of manager
changes on fund flows. We create a dummy variable, FemNewi,t−1 (MgrChgi,t−1), which is equal to
one if a male fund manager is replaced by a female fund manager (if any manager change occurs), and
zero otherwise. The results show that fund flows decrease by about 13% if a male manager is replaced
by a female manager, while a manager change per se has no significant impact.
Another possible explanation for the low inflows into female-managed funds could be that female
and male fund managers differ with respect to other demographic characteristics that investors might
consider in their investment decision. Results from panel B in Table 1 show that male and female
managers indeed differ, e.g., with respect to their tenure at a particular fund, their age, and the
probability that they hold a PhD degree. Thus, in Column 2, we add further control variables that
capture the impact of these differences on flows. We did not include these variables in our base model,
because we only have information on the demographic characteristics for a subset of fund managers.
We include dummy variables that take on the value one if the manager holds a MBA degree, a PhD,
or a professional qualification (e.g., CFA), respectively, and zero otherwise, as well as a fund manager’s
age and tenure at the fund currently managed.10 We find that manager tenure has a positive impact
on fund flows, while age and education have no significant impact. Irrespective of this, we still find
that female managers receive on average nearly 12% lower inflows after adding these additional control
variables.11
Kaniel, Starks, and Vasudevan (2007) show that media coverage can have a positive impact on fund
flows. A similar effect is documented for fund advertising in Jain and Wu (2000), Cronqvist (2006),
and Gallaher, Kaniel, and Starks (2015). The results from panel B in Table 1 show that the press
covers male fund managers significantly more often than female managers, while 12b-1 fees (which
are explicitly labeled to cover distribution and marketing expenses) are higher for female-managed
funds. To control for the impact of these differences, in Column 3, we thus add lagged media coverage,
10We do not include a separate dummy for Bachelor degrees, as virtually all managers hold at least a Bachelor’s degree(see panel A in Table 1). Some fund managers hold Masters degrees other than MBAs. Including controls for non-MBAMasters does not change our findings.
11To test whether the negative impact of gender might be weakened if the fund manager has a good education or a longtenure, in unreported tests, we also interact the impact of the female dummy with the education dummies and an abovemedian tenure dummy. In all cases, we find no significant impact of the interaction terms, while our main results remainunaffected.
9
LN(1+MedCov)i,t−1, defined as the natural logarithm of the number of articles on fund i’s manager in
year t−1, as an additional control variable. Results show that media coverage does have a significantly
positive impact on fund flows. However, including media coverage does not significantly change the
coefficient of our female dummy.12 In Column 4, we include 12b-1 fees as a proxy for advertising and
other marketing expenditures which again does not change our main result.
It is also possible that it is not gender bias of investors themselves that drives our results, but
that brokers who advise investors steer them away from female-managed funds. There is some indirect
evidence suggesting that fund brokers might stereotype women as less competent in financial matters
and might thus promote male-managed funds more often than female-managed funds. For example, a
survey conducted by Wang (1994) suggests some ’machismo’ among brokers: sales representatives at
brokerages spend more time advising men than women, offer a wider variety of investments to men,
and try harder to acquire men as customers. Thus, in Column 5, we investigate whether the negative
impact of our female dummy on mutual fund flows is driven by funds that are distributed via brokers.
As such funds typically charge front-end loads (Christoffersen, Evans, and Musto (2013)), we interact
our female dummy with a dummy variable which is equal to one if none of the fund’s share classes
charges a front-end load, and zero otherwise. We do not find a significant difference between no-load
funds and load funds suggesting that the negative impact of our female dummy on mutual fund flows
is not driven by brokers.
Another concern is that our results are not really due to investors preferring male fund managers,
but can be explained by male managers having better access to often male-dominated networks of
institutional investors. Thus, we also run our regression separately on a subsample of funds that only
offer retail share classes and on a subsample of funds that only offer institutional share classes. Results
presented in Columns 6 and 7 show that the effect of the female dummy is of similar economic magnitude
and even slightly larger among funds focusing on retail investors exclusively. It is insignificant among
institutional funds (probably due to the small number of observations), but significant at the 1% level
among retail funds.
12In unreported tests, we also add an interaction term between media coverage and the female manager dummy. Theinteraction term is not significant (indicating that press coverage on females is neither more positive nor particularlynegative as compared to that of male managers). Again, our main result remain unaffected.
10
3.3 Robustness
We now analyze whether our results are robust to further variations of our empirical strategy. In panel
B of Table 3 we present results for modifications of our base model (Column 3 in panel A of Table
3). The same controls are included in the estimation but suppressed in the table. We start by using
alternative measures of fund flows as described in Appendix B. First, in Column 1, we use dollar flows,
AbsF lowi,t, instead of relative flows as dependent variable. We still find a significantly negative impact
of the female dummy variable on fund flows that is also economically meaningful: a female-managed
fund on average gets about 14.3 million USD less money inflows p.a. than a comparable male-managed
fund. This translates into female-managed funds growing by about 19.5% less than male-managed
funds. Second, in Column 2, we follow Spiegel and Zhang (2013) and use the change of a fund’s
market share, ChgMktShri,t, as dependent variable, exclude the lagged dependent variable from our
regression and estimate the model using quantile regressions.13 As in Spiegel and Zhang (2013), we
now do not find much evidence for a significantly convex performance-flow relationship anymore (the
squared performance rank is only marginally significant). However, the female dummy variable is still
significantly negative. Third, in Column 3, we use monthly instead of yearly relative flows as dependent
variable and run our regressions on a monthly basis. In this regression, we only include those controls
that also change on a monthly basis. We again find a highly statistically significant negative coefficient
indicating that female-managed funds grow by about 5% p.a. less than male-managed funds.
In our previous analysis, we use a quadratic specification to model the impact of past performance
on inflows. As an alternative specification, we estimate a piecewise linear relationship in Columns 4
and 5. Specifically, we estimate distinct slope coefficients for different performance quintiles.14 Results
based on return ranks and Carhart (1997) four factor alpha ranks confirm the convex performance-flow
relationship and we still find a negative impact of the female dummy.
To assess the temporal stability of our findings, we split up our sample into two time periods, up
to 2001 and after 2001, as well as into years with negative market returns (2000, 2001, 2002, and 2008)
and years with positive market returns (all other sample years). Results presented in Columns 6 to 9
show a significantly negative impact of the female fund manager variable in all cases. The effect is even
somewhat stronger in later years, but there is no big difference between good and bad market years.
13In Spiegel and Zhang (2013) the authors use vigintiles in their analysis. However, there are not always observationsfor female managers in each vigintile and year. Thus, we use quintiles instead of vigintiles.
14We follow Sirri and Tufano (1998) by grouping the three middle quintiles together. Results (not reported) do notchange if we model a distinct slope coefficient for each of the five performance quintiles separately instead of grouping thethree middle quintiles together.
11
Finally, in panel C of Table 3, we present results from a matched sample analysis. For each obser-
vation with a female manager we try to find male-managed twin funds with similar characteristics. We
use different combinations of matching criteria. In all cases, we require the matching observations to
be from the same year and segment and to be in the same size decile in the respective year. We always
match based on fund size because this variable has the strongest and most consistent influence on flows
in Table 2. Then, we re-run the same regression as in Column 3 in panel A of Table 2 based on our
matched sample. Results in Column 1 of Table 3 show a highly significant negative impact of the female
dummy on flows which amounts to 7.5% p.a. In Columns 2 to 6 we additionally require the matching
funds to be in the same fund-age decile, manager-age decile, lagged return rank decile, have a manager
with the same level of eduction, and to be in the same manager fund tenure decile, respectively. The
results show a very uniform picture: the impact of the female dummy is always significantly negative
and economically meaningful, ranging from -8% to -10% p.a.
4 Gender bias vs. rational statistical discrimination
Results in the previous section suggest that investors prefer male-managed funds to female-managed
funds. We propose gender bias as one possible explanation for this finding. However, our findings could
also be driven by statistical discrimination rather than by a gender bias. To disentangle these two
explanations, we now investigate whether there is any evidence of undesirable investment behavior
(Section 4.1) or inferior fund performance (Section 4.2) of female fund managers as compared to male
fund managers.
4.1 Investment styles
It is sometimes argued that gender differences are of little importance among professionals, because the
similar environment and educational background of professionals overrides potential gender differences.
However, there is also evidence that gender differences are relevant in professional management settings
(e.g., Adams and Funk (2012) and Graham, Harvey, and Puri (2013)).
To examine gender differences between male and female fund managers, we relate various measures
of investment behavior to the fund manager’s gender and other potentially relevant fund characteristics.
We focus on risk-taking behavior, trading activity, and the variability of investment styles over time.
12
In our regressions, we either use one of three risk measures for fund i in year t, FundRiski,t,
SysRiski,t, or UnsysRiski,t, or the fund’s turnover ratio, TORatioi,t, all as defined in Appendix B,
as dependent variable. Besides the female manager dummy, we include fund size and age as control
variables. Furthermore, we include a fund’s previous year return, FundReti,t−1, the fund manager’s
tenure, MgrTenurei,t−1, as well as time, segment, and fund company fixed effects. We include segment
and fund company fixed effects because some segments are more risky than others and because man-
agement companies often have specific guidelines or cultures in place that can have a strong impact on
a fund manager’s investment behavior. Including fund company fixed effects also addresses the concern
that female managers might self-select into low-risk fund companies. Standard errors are clustered at
the fund level. Panel A of Table 4 summarizes our findings.
Regarding the various dimensions of risk taking behavior, we find negative coefficients for the impact
of a female manager, which is consistent with the widely documented fact that women tend to be more
risk-averse (e.g., Byrnes, Miller, and Schafer (1999)). We also find that women tend to trade less,
which is often interpreted as evidence for less overconfidence (Barber and Odean (2001)). However,
both effects are not statistically significant.
Finally, we want to examine whether there are any differences in style variability defined as the
variability of a fund’s factor loadings over time (see Appendix B).15 We only conduct a univariate com-
parison between the style variability measures of female- and male-managed funds, because we only
calculate one style variability measure based on the entire time span over which a specific manager
manages a fund. Results in panel B show that style variability is significantly lower for female-managed
funds, i.e., female fund managers follow more stable investment styles over time than male fund man-
agers. This finding holds for the overall style variability measure (Column 1) as well as for the three
factor individual style variability measures (Columns 2 to 4).16
Overall, we find only minor differences with respect to the investment behavior of female and male
fund managers: female fund managers’ investment behavior should be, ceteris paribus, more desirable
for mutual fund investors as they follow more stable and thus reliable investment styles than male fund
managers.
15In unreported tests we also compare average factor loadings and find that women tend to have significantly lower(higher) loadings on the HML (MOM) factor, while there is no significant difference with respect to SMB loadings.
16Estimates of standard deviations can be biased if they are based on a small number of observations. Thus, we repeatour analysis using the variance of factor loadings over time. Results (not reported) are qualitatively similar, but significanceslightly decreases.
13
4.2 Fund performance
We now examine whether the behavioral differences documented in the previous section have an impact
on fund performance and performance persistence.17 We start by relating various performance measures
of fund i in year t to a female dummy and controls. As performance measures we use a fund’s yearly
net return, its one-, three- and four-factor Alpha, its Sharpe-Ratio, and an extended version of the
Appraisal Ratio of Treynor and Black (1973), all as defined in Appendix B. Results based on panel
regressions with time, segment, and fund company fixed effects as well as standard errors clustered at
the fund level are presented in panel A of Table 5.
There is no significant difference between the performance of female- and male-managed funds.
This result holds irrespective of the specific performance measure we use. Panel B presents results of
various further robustness tests. We only present the coefficient estimate for the impact of the female
dummy, but the same controls as above are included. In line B.1 we add additional fund characteristics
as controls. Specifically, we include the fund’s lagged inflows to control for the effect that it might be
more difficult for male managers to perform well because they face larger inflows. Other additional fund
characteristics we include are the fund’s lagged turnover ratio, lagged performance, and lagged fund
risk. In line B.2 we include variables capturing the influence of the manager’s age and dummy variables
reflecting the manager’s education (MBA, PhD, professional qualification). In both cases and for all
performance measures we still can confirm that there is no significant performance difference between
male and female fund managers. This also holds if we estimate the same models as in panel A, but run
Fama and MacBeth (1973) regressions (B.3). Finally, we conduct a matched sample analysis, where
each female-managed fund year observation is matched with male-managed fund year observations from
the same segment, the same year, and the same size decile (B.4). We specifically match based on size
to control for the potential impact of diseconomies of scale. Again, there is no significant influence of
the female dummy on any of the performance measures.18
As individual fund performance can only be estimated with noise we also analyze the performance of
equal-weighted portfolios consisting of female- and male-managed funds, respectively, as an alternative
to the multivariate regression approach. We evaluate the performance of a hypothetical difference
portfolio that is long in all female-managed funds and short in all male-managed funds. Results are
17e.g., Brown, Harlow, and Zhang (2014) document a positive influence of stable investment styles on performance.18Glode (2011) argues that investors particularly value good performance during bad states of the economy. Thus, we
also check whether males might deliver better returns than females during market downturns. In unreported tests we findno difference in the impact of the female dummy on various performance measures across market states.
14
presented in panel C. Irrespective of whether we focus on Jensen (1968) one-factor Alphas, Fama and
French (1993) three-factor Alphas, or Carhart (1997) four-factor Alphas, the difference portfolio never
delivers any statistically significant abnormal returns.19
Taken together, our results suggest that the market for mutual fund managers is efficient in the
sense that it is not possible to generate abnormal returns by following an investment strategy based
on a manager characteristic as easily observable as the manager’s gender. Although female and male
fund managers differ in terms of investment behavior, these differences are not reflected in differences
in average fund performance.
In panel D we analyze gender differences in performance persistence. Performance persistence is
defined as the standard deviation of a manager’s performance ranks over time.20 We investigate per-
formance persistence based on the five performance measures analyzed above. Results show that the
performance ranks of male-managed funds are more variable over time than those of female-managed
funds. The effect is statistically significant for most performance measures. This provides some ev-
idence that the performance of female-managed funds is more persistent than the performance of
male-managed funds. A more stable performance as well as the more stable investment styles of female
managers documented above should, if anything, be preferable from an investor’s point of view.
Overall, the evidence provided in this section is not consistent with the idea of investors rationally
avoiding female fund managers. Rather, it suggests that investors exhibit taste-based irrational behavior
leading to gender bias.
5 Do investors care about the manager’s gender? - Experimental
evidence
Although the previous sections suggest that rational statistical discrimination and several other al-
ternative explanations for lower inflows into female-managed funds are unlikely to be the main driver
of our results, it is of course not possible to empirically observe and control for all other potential
drivers of fund flows. Thus, to shed further light on the question whether investors really care about
manager gender, we conduct a controlled laboratory experiment to better identify a causal impact of
fund manager gender on flows. This procedure also has the advantage that we can examine the impact
19In unreported tests, we also analyze value-weighted portfolios. Results are again insignificant.20Analyzing a variance based measure of performance persistence delivers qualitatively identical results.
15
of investor characteristics on investment decisions, while the previous empirical analysis focusses on
aggregate investor behavior at the fund level.
The experiment was conducted with U.S. university students and consists of two main parts, an
investment task (Section 5.1) and an Implicit Association Test (IAT, see Section 5.2). The investment
task experiment allows us to analyze the impact of manager gender on capital allocations in a controlled
laboratory setting and the IAT allows us to get a proxy for gender bias on an individual level. We then
link back IAT scores to investment decisions to test whether gender bias and investment behavior are
linked (Section 5.3). Details of the experimental procedure are described in Appendix D.
5.1 Investment task
We develop a simple between-subjects design in which 100 experimental currency units have to be split
between two S&P 500 index funds that we randomly chose from the CRSP fund database beforehand.
Since index funds barely differ from each other, they offer the cleanest setting in which to examine the
impact of specific variables on investment decisions (Choi, Laibson, and Madrian (2011)).
In each investment round, the complete amount of 100 experimental units has to be invested. Instead
of providing the funds’ real names, we labeled them “Fund A” and “Fund B” to avoid any framing
or familiarity effects. At the beginning of each investment round, information about both funds was
displayed to subjects and they subsequently decided how to split their money between those funds.
Subjects were randomly assigned to one of two groups, group X or group Y. Both groups were shown
information on the funds. However, we manipulated the gender of the fund manager between these
groups, while keeping all other information constant. Figure 3 shows the information given to the two
groups of subjects. The only difference between both groups of subjects is the first name of the fund
manager. Group X observes a female fund manager for fund A and a male fund manager for fund
B, while group Y observes a male fund manager for fund A and a female fund manager for fund B,
respectively.21 This procedure allows us to attribute any differences in investment behavior between
the two groups solely to the fund manager’s gender.
The experiment consisted of four rounds.22 Investment rounds only differed with respect to the
amount of information provided about the funds. In the first round, information about the fund segment,
21We took the most common U.S. first names according to the U.S. Social Security Administration to ensure thatsubjects perceive these names as very common for each gender category and we use common last names.
22The experiment was part of a more extensive investigation where subjects also made additional investment decisions.In this paper we only report the results relevant in our context, i.e., the impact of gender on index fund investments.
16
the name of the fund manager, fund size, inception date, expense ratio, trading activity, and top five
stock holdings was provided. In addition, we added a short text labeled “Fund Facts” with a description
of the fund’s investment strategy (see Figure 3). In the following three rounds we added additional
information: an ethical rating of the fund, a classification indicating the fund’s riskiness, and the fund’s
return over the past 12 and 24 months, respectively.
We recruited 100 students as subjects in our experiment. Table 6 provides information on the
demographic characteristics of the subjects. Due to the recruiting procedure (about 50% of the an-
nouncements were made in finance classes) most subjects (i.e., 43 individuals) indicated “Finance” as
their main field of study, followed by 13 subjects in “Accounting”, 10 in “Marketing”, and 9 in “Man-
agement Information Systems”. A smaller number of subjects indicated “Economics”, “Engineering”,
or other fields as their main field of study. The mean age of subjects is 21.3 years and ranges from
a minimum of 18 years to a maximum of 40 years. Virtually all subjects were single and the gender
distribution is roughly balanced, with 51 male and 49 female subjects. Results from the investment
task are reported in Table 7.
In our setting, we compare differences in the amount invested in fund A between group X (which
observed a female manager of fund A) and group Y (which observed a male manager of fund A) to
isolate the impact of the fund manager’s gender on investment behavior. Panel A of Table 7 presents
results based on all four rounds in the first line. Strictly speaking, only the first round of investment
decisions can be considered to be completely independent in an experiment like ours, where subsequent
rounds involve investment choices regarding the same pair of funds. Thus, in the second line we focus
on the first round of the experiment only.
In both cases, results show that subjects generally invest less into fund A as compared to fund B
(i.e., in both groups the fraction invested is below 50%) which might be due to fund A’s higher expense
ratio (see Figure 3). However, although fees should be the only consideration in choosing between
index funds and the whole amount should be invested in the cheaper fund, we find that subjects invest
significant amounts in both funds. This finding confirms results from a similar experiment reported in
Choi, Laibson, and Madrian (2011).
More important in our context, subjects invest significantly less in fund A if it is managed by a
female fund manager than if it is managed by a male fund manager.23 The difference is 7.42 experimental
23Note, that we only compare investments in fund A between subjects conditional on the fund manager’s gender. Thus,the amounts shown in Column 1 and 2 in Table 7 do not add up to 100. By definition, our conclusions would remainunchanged if we compared investments in fund B instead.
17
units or roughly 15% and is significant at the 1% level if we pool observations from all rounds. It is
even larger (8.51 experimental units) and significant at the 5% level if we only focus on the first round
investment decisions.
In the following panels, we split up observations by various subject characteristics. In order to
prevent samples from getting too small, we focus on results based on observations from all rounds. In
panel B, we split up subjects by gender. Results show that the difference in investing in female- and
male-managed funds is mainly driven by male subjects. We find no significant difference in the fraction
of money invested between male- and female-managed funds among female subjects. Panel C shows that
the bias towards male-managed funds is independent of the main field of study of the subjects. Panel
D splits the subject pool by financial literacy. We observe significantly less money directed towards
the female-managed fund in both groups, but the effect seems to be slightly stronger among the more
financially literate. Furthermore, as one would hope, the more financial literate subjects seem to be
more sensitive to fund expenses. On average, they invest only 35.3 experimental units in the more
expensive index fund, as compared to 46.6 experimental units invested in this fund by the less financial
literate subjects.
Overall, our experimental evidence confirms the empirical evidence from Section 3. As all other
potential drivers of fund flows are controlled for in this setting, these results suggest that our previous
empirical findings are indeed due to the managers’ gender and support our conjecture of investors
preferring male-managed funds.
5.2 Implicit Association Test
In the second part of the experiment, we conducted an implicit association test (IAT) to directly
examine whether gender bias explains the observed investment behavior in the laboratory. The IAT
has gained enormous popularity among social psychologists in recent years as it can uncover prejudice
based on simple associations. According to Lane, Banaji, Nosek, and Greenwald (2007), there are
now well over 200 papers that use this method. In previous applications, the IAT is used to uncover
various social biases like prejudice against different races, religions, genders, or sexual orientations.
The test’s popularity is based on the fact that it can be easily administered and that it allows to
also uncover implicit prejudice that subjects are often not willing to admit openly because of social
desirability concerns. Even if complete anonymity is credibly guaranteed, respondents often do not
answer truthfully in standard surveys. In contrast, the IAT provides a simple way to measure prejudice
18
based on automatically operating implicit associations that cannot be easily manipulated and might
even operate completely unconsciously (Greenwald, Banaji, Rudman, Farnham, Nosek, and Mellott
(2002)). Its reliability and validity is widely confirmed by showing that IAT scores predict biased
behavior in many contexts like voting behavior or brand choices (Cunningham, Preacher, and Banaji
(2001), Greenwald, Poehlman, Uhlmann, and Banaji (2009)).
As IAT tests are not used in the finance literature so far, we will now shortly describe how a
typical gender IAT works: Subjects are required to classify items into one of four categories (e.g.,
‘Male’ or ‘Female’ and ‘Science’ or ‘Liberal Arts’) in a computerized double-sorting task. Two of the
four categories are displayed on the left side of the screen, while the other two are displayed on the
right side of the screen. In the ’stereotypical’ or compatible configuration, ’Male’ and ’Science’ would
be displayed together on one side and ‘Female’ and ‘Liberal Arts’ would be displayed together on the
other side, while in the incompatible configuration one of the categories is switched from one side of
the screen to the other (e.g., ‘Female’ and ‘Science’ would show up on the same side). Subjects have
to rapidly sort items appearing in the middle of the screen by hitting either a left- or a right-hand key.
The IAT measures reaction times in the two configurations. The test relies on the fact that stronger
associations (e.g., ‘Male’ with ‘Science’) result in faster reaction times than weaker associations (e.g.,
‘Female’ with ‘Science’) and that the strength of associations serves as a proxy for implicit prejudice.
If there is no implicit prejudice, average reaction times should be identical. In contrast, if there is a
biased perception that, e.g., men are more skilled in science and women are more skilled in liberal arts,
reaction times would be higher in the incompatible configuration.
To examine whether there is any evidence of gender bias in our setting, we adapt the IAT to
the context of finance. The first category we use is ‘Male’ vs. ‘Female’. The words belonging to the
gender categories are taken from typical gender IATs like the one described above. They are all easily
recognizable as belonging to the female or male category like ‘father’, ‘uncle’, ‘mother’, or ‘aunt’. The
full list of items is presented in panel A of Table 8. The second category we use is ‘Finance’ and
‘Marketing’. We chose ’Marketing’ as the contrasting category, because finance and marketing are two
of the most prominent majors among U.S. undergraduate students in business administration. The
items that have to be sorted into these categories are again easily recognizable and include ‘stocks’,
‘mutual funds’, ‘advertising’, and ‘logo’. The full list of items is presented in panel B of Table 8. Subjects
have to categorize items by hitting the ‘E’ or ‘I’ key on their keyboards, depending on whether the
19
specific item displayed on the center of the screen belongs to a category displayed on the left-hand or
right-hand side of the screen. An example is provided in Figure 4.
Panel A displays the compatible configuration where the categories ‘Finance’ and ‘Male’ are on one
side of the screen and ‘Marketing’ and ‘Female’ are on the other side. In contrast, panel B displays the
incompatible configuration. In both cases of the example shown in Figure 4, subjects had to sort the
item ‘stocks’ into the right category as fast as possible. If their reaction time is significantly higher in
the incompatible configuration than in the compatible configuration, this indicates that they are more
biased. The test was administered in two versions and subjects were randomly assigned to one of the
versions. Subjects assigned to the first version of the test started with the compatible configuration
followed by the incompatible configuration, and vice versa for subjects assigned to the second version.
After several practice rounds, in which subjects could get familiar with the sorting task, we start
measuring their reaction times.
The simplest way to compute IAT scores is to just compare reaction times in milliseconds (ms),
which we denote by R. The reaction times for both groups in the compatible and the incompatible con-
figuration are summarized in box-plots presented in Figure 5.24 Panel A (B) reports results for subjects
who first played the compatible (incompatible) configuration and then the incompatible (compatible)
configuration. In both cases, reaction times are lower in the compatible than in the incompatible config-
uration. In panel A (B), the mean reaction time for the compatible configuration is 753.99 ms (833.13
ms), while it is 914.15 ms (994.79 ms) in the incompatible configuration. To examine reaction times
more formally we aggregate data on the subject level and calculate the average reaction time using three
alternative methods. First, we compute the simple average of the reaction times R in ms. This approach
has the advantage that effects can be directly interpreted. Second, we calculate log-transformed reac-
tion times, log(R). This approach has the advantage that the distribution of log-transformed reaction
times has a more stable variance and is thus more suitable for statistical analysis. Third, we calculate
a speed variable defined as S = 1,000R . This variable also has desirable distributional characteristics
that stabilize variances and can be directly interpreted as items per second. We then calculate the
corresponding IAT score as the difference in the mean reaction time between the compatible and the
incompatible configuration based on R, log(R), and S for each subject j. These scores are suggested
in Greenwald, McGhee, and Schwartz (1998) and are denoted by d(R)j , d(log(R))j , and d(S)j , respec-
24To prevent outliers from driving the results we follow Greenwald, McGhee, and Schwartz (1998) and set all unreal-istically long reactions times (over 3 seconds) equal to 3 seconds and all unrealistically short reaction times (below 300ms) equal to 300 ms.
20
tively.25 Independent of the configuration a subject plays first, we always subtract the mean reaction
time in the compatible configuration from the reaction time in the incompatible configuration for R
and log(R), and vice versa for S. Thus, if d is significantly larger than zero, this suggests the existence
of a gender bias.
Results for a pooled examination of all subjects are presented in Table 9. The mean of d(R) across all
subjects is 160.96 ms, i.e., the average of the subject individual mean reaction times in the incompatible
configuration is 160.96 ms or about 18% higher than in the compatible configuration. The hypothesis
that the IAT score is not different from zero can be rejected at the 1% level (t-statistic > 10). This also
holds for the other measures d(log(R)) and d(S). In the last four columns, we present the number and
percentage of subjects for which the respective d measure is (at least at the 10% level) significantly
negative, negative, positive, and (at least at the 10% level) significantly positive, respectively, on an
individual level. 62% of the subjects show a significantly positive d even on an individual level. Only
4% exhibit a significantly negative d. These results provide evidence that most of our subjects indeed
show signs of gender bias in a financial context.
We also investigate which subject characteristics are related to the strength of gender bias. We first
compare male and female subjects as well as finance and marketing students. Results are presented in
panels A and B of Table 10 and show significant gender bias among all groups. The differences between
the groups are not statistically significant.
Tajfel (1970) provides evidence for an in-group bias of individuals. This effect should lead to less
pronounced or no gender bias among female finance students. In panel C, we find that the 25 male
subjects that study finance show an average difference in reaction times of 224 ms, which is clearly
larger than the typically observed effect of about 160 ms in the overall subject population. In contrast,
among the 18 female subjects who study finance, the difference amounts to only 118 ms. Interestingly,
this effect is still significant at the 5% level, but is only about half the size of the effect observed
among male finance students. Moreover, the difference between male and female finance students is
also statistically significant (t-statistic: 2.05, based on d(R)).
25Alternatively, we use the pooled standard deviation from both configurations as effect size unit to get subject-
individual adjusted measures dadj for implicit prejudice. For example, dadj(R) is defined as dadj(R) = RI−RC
std(R), where RC
(RI) denotes mean trial reaction times from the compatible (incompatible) configuration, and std(R) denotes the pooledstandard deviation of reaction times from both configurations. These measures, for which the variance is more stable,allow us to detect statistical effects more precisely. Results (not reported) using these adjusted measures are very similar.
21
Finally, in panel D we check whether there is any relation between the level of financial literacy
and the IAT score. The average IAT score in the high financial literacy group is 177 ms vs. 150 ms in
the low financial literacy group but the difference is not statistically significant.
Results in experiments often crucially depend upon the experimental procedure. Thus, we also
test whether the results are stable against variations of the experimental parameters. Specifically, in
panels E to G we check whether results depend upon the gender of the instructor in the experiment,
on the time of the day (Folkard (1976)), or on differences in the number of subjects per session, i.e.,
the crowdedness of the sessions (Paulus, Annis, Seta, Schkade, and Matthews (1976)). Our results are
unaffected by these parameters.
5.3 Impact of investor-level IAT scores on investment decisions
Overall, the results from the IAT are consistent with the view that there is gender bias in finance.
However, it is unclear whether this bias affects investment behavior and is strong enough to eventually
result in lower inflows into female-managed funds. Thus, we now compare the fraction invested in
female-managed funds in the investment task of the experiment between subjects with a strong gender
bias according to their IAT score to subjects with no or even a reverse gender bias. Results are presented
in Table 11.
Panel A shows the mean amounts invested in the male- and female-managed index funds over all
rounds separating between subjects with high and low IAT scores. The results show that subjects with
high IAT scores (d(R) > 0, d(log(R)) > 0, d(S) > 0) invest significantly less in female-managed funds.
In contrast, we find (insignificantly) larger investments in female-managed funds of those subjects with
negative IAT scores.
In panel B, we present multivariate evidence from a censored Tobit regression with the fraction
of experimental units invested in index fund A—which can either have a male manager (group X)
or a female manager (group Y)—by subject j as dependent variable. As independent variables we
include a female manager dummy, that takes on the value 1 if fund A as presented to subject j is
managed by a female, and zero otherwise, as well as a set of control variables. We include (but do not
explicitly report for the sake of brevity) dummies that take on the value one, if subject j has an above
median IAT score, (SubjIATj), is female (SubjGenj), studies finance or economics (FinEconj), has
above median financial literacy (HighFinLitj), faced a female instructor explaining the experiment
22
(InstrGenj), is married (SubjMaritalj), and has investment experience (EverInvestj), respectively,
and zero otherwise, as well as the age of the subject in years (SubjAgej). Regressions are estimated
with session fixed effects.
Results in Column 1 confirm our earlier results from Table 7 and show that fund A receives 9.3
experimental units or nearly 20% less if it has a female manager. More interestingly, in Column 2 we
interact the female manager dummy with a dummy equal to one if a subject showed above median IAT
scores, and zero otherwise. The interaction term is significantly negative. The coefficient indicates that
subjects with above median IAT scores on average allocate 17.3 experimental units less to fund A if it is
managed by a female manager as compared to the base case. The linear impact of the female manager
dummy itself is now insignificant. This result confirms our earlier univariate finding from panel A.
In Column 3, we add an interaction term between the female manager dummy and the female
subject dummy. The coefficient on the interaction term is significantly positive and nearly as large as
the impact of the female manager dummy itself. This confirms our earlier findings from panel B in
Table 10 and shows that the negative impact of a female manager is neutralized if the subject making
the investment decision is female. In Columns 4 to 6 we interact the female dummy with a dummy
for finance/economics students, with a dummy for high financial literacy, and with a female instructor
dummy, respectively. None of these interaction terms is significant.
Overall, the results from the experiment suggest that many individuals are subject to gender bias
as measured by IAT scores and that this bias has a very strong impact on investment decisions.
6 Discussion and equilibrium implications
In this section, we discuss some potential remaining concerns regarding our results as well as the
equilibrium implications of our findings. First, we address the concern that investors might not even
know who the fund manager is (Section 6.1). Second, in Section 6.2, we discuss whether our results
could be consistent with a rational equilibrium as described in Berk and Green (2004). Finally, we
turn to the question why we see any women at all in the mutual fund industry, given that they attract
significantly lower inflows (Section 6.3).
23
6.1 Do investors know who manages their fund?
One might be concerned about whether fund investors are aware of who is managing the fund they
invest in. First, it is important to note that it does not matter so much for our analysis whether
investors remember who manages their fund at a later point in time. It is only important that investors
are exposed to the identity of the manager when they make their investment decision. The literature
on social categorization processes has shown that social biases are automatically activated by the mere
presence of a stimulus. With respect to gender as a social category, several papers have shown that
exposure to information about gender, as conveyed through names, pictures, or gender stereotypical
words, can exert an unconscious influence on individual decision making (Banaji and Greenwald (1995),
Blair and Banaji (1996)). Thus, even if mutual fund investment decisions do not consciously rely on
the gender of a fund manager, they can be influenced by investors’ perception of the manager’s name,
particularly if the name evokes any unconscious stereotypes or other emotional responses.
Second, we can show that information on the fund manager is usually easily available to investors:
We collect fund information for the largest single-managed fund of the fifty largest fund companies
in our sample. Out of these funds, 98% report the fund manager’s name online on their webpage as
well as in the official prospectus.26 Furthermore, besides prospectuses and fund company websites,
many investors rely on financial websites like, e.g., Yahoo Finance to gather fund information. Figure
2 presents screenshots of the information investors would get if they search for a specific fund in four
of the major online financial information sources. As can be seen from these exhibits, information on
the gender of the fund manager is salient to investors as it can typically be easily inferred from the
first name of the fund manager, which is prominently presented on the first page that appears.
Additional evidence that investors are often directly exposed to manager names comes from prod-
uct descriptions in personal finance magazines. For example, Kiplinger Magazine—one of the leading
personal finance magazines in the U.S.—features a Top 25 list (KIP25) of funds on its webpage. For
many funds, a short feature article appears if investors click on the fund name. For example, there were
articles available for 11 of the 15 U.S. equity funds contained in the list (in November 2011). Eight out
of those eleven articles mentioned the name of the fund manager in the very first sentence.
26The only two companies that did not report the manager’s name on their webpage is Dimensional Fund Advisory andCapital Growth Management. However, even these companies reported the manager’s name in the prospectus. For 74%of the funds, the manager name was reported on the main website of the fund (instead of only being visible after clickingonce more or just being included in the fund prospectus).
24
Finally, evidence that fund managers’ identities matter for investment decisions of mutual fund
investors is also provided in earlier empirical papers on mutual fund flows. For example, Massa, Reuter,
and Zitzewitz (2010) show that funds have greater inflows if the name of the fund manager is declared
as compared to funds where the manager name is kept anonymous. They also show that departures
of named managers reduce inflows. Furthermore, Kumar, Niessen-Ruenzi, and Spalt (2015) show that
fund investors shy away from funds with managers with foreign sounding names and that this effect
got stronger for Middle Eastern sounding names after 9/11. These results suggest that a sufficiently
large fraction of investors takes the manager’s name into account.
From the evidence provided in this section, we conclude that manager information is generally
available to investors and that investors are often exposed to and take into account fund manager
names when making investment decisions.
6.2 Lower inflows in a rational equilibrium?
Berk and Green (2004) show theoretically that the observed performance of all fund managers is
identical in equilibrium even if their skill levels differ. The reason for this result is that they assume
that fund managers’ investment skills are subject to decreasing returns to scale. If there is competitive
provision of capital by investors in the form of money inflows, this leads to an equilibrium where all
funds grew to a size at which they are not able to outperform any longer. In a perfect Berk and
Green (2004) world, investors might rationally predict that female fund managers would underperform
if they received larger inflows. Thus, they provide less capital to female fund managers. However,
recent empirical evidence questions the underlying assumption of the Berk and Green (2004) model
that there are strong diseconomies of scale in the fund industry (see Reuter and Zitzewitz (2013)).
Furthermore, our results obtained from the controlled laboratory experiment in Section 5.1 clearly
cannot be explained by the Berk and Green (2004) model. The findings reported there are based on
investment decisions between a female and a male-managed index fund. One reason why we focused
on index funds is that the ability of the manager to outperform the market is irrelevant for this type
of fund. In addition, Chen, Hong, Huang, and Kubik (2004) argue that diseconomies of scale are not
important for index funds. Consequently, the Berk and Green (2004) equilibrium argument is not
relevant in this context. We conclude that it is unlikely that the flow effects we document using field
data and particularly the experimental evidence can be explained as a rational equilibrium response of
investors as described in Berk and Green (2004).
25
6.3 Why not even fewer female fund managers?
One provocative question that one may ask based on our findings is why we observe any female fund
managers at all. One could argue that it is suboptimal for fund management companies to employ
female fund managers at all if they attract lower inflows than male managers. However, while our
results show that investors on average shy away from female-managed funds, this does not mean that
all investors behave like this. If there is a significant number of investors who do not display a gender
bias (or even actively favor females) it makes sense for fund companies to employ at least some female
managers to cater to those investors.
Results from the experimental investment task show that there is a minority of subjects (typically
women) who are not biased against female fund managers or even invest more with them. Therefore, it
can still make sense from the fund company’s point of view to hire female fund managers to specifically
cater to this group of investors.27 Furthermore, many institutional investors require their business
partners to report explicitly on their diversity policy. In a similar vein, the Dodd-Frank Act requires
federal agencies to do business only with firms that “ensure the fair inclusion of women” and to “give
consideration to the diversity of the applicant” (Dodd-Frank Financial Regulation Bill Section 342(c)).
For mutual fund companies to win mandates from such clients, it is necessary to employ at least some
female fund managers.
However, most regulations and diversity policies of institutional investors do not prescribe them to
invest in female-managed funds. Rather, they typically only have to make sure that the companies they
do business with have some diversity policy in place. Thus, it could be the case that fund companies
employ some female managers to formally fulfill the requests of such investors. However, these investors
might still not invest in the female-managed funds, but rather in the other funds of the company. Then,
female fund managers would not directly attract flows into their own funds, but their presence in the
company would lead to positive spill-over effects for the other funds of the company.
To test this idea, we adapt the flow regression from Column 3 in Table 2 to capture such potential
spill-over effects and run the regression based on male-managed funds only. Results are presented in
Table 12. In Column 1, we replace the female dummy by a dummy variable taking on the value one, if
there is another female-managed fund among the single-managed funds of the same fund company, and
27Consistent with this argument, there are indeed some niche funds like the Pax World Global Women’s Equality Fundthat specifically cater to female investors.
26
zero otherwise.28 In Column 2, instead of the dummy variable, we use the number of female-managed
funds in the same company as independent variable. In both cases, we find a highly significant positive
impact of the spill-over variable. For example, the coefficient in Column 1 indicates that male-managed
funds grow by more than 6% p.a. more if the fund company also employs at least one female manager.
This seems like a very large impact and gives rise to the question whether fund companies should not
employ more female managers in order to profit from these large indirect positive flow effects. However,
in Column 3, we present results where we include both spill-over variables simultaneously. We find a
highly significant impact of the variable indicating the presence of at least one female-managed fund,
while the number of female managers now is insignificant, i.e., there seems to be no additional benefit
of adding more female managers if there is already at least one female-managed fund in the company.
Overall, our results from Table 12 are consistent with the argument that fund companies should at
least employ some female fund managers despite the lower inflows they generate because of the demand
of certain investor groups requiring them to document the inclusion of women. The observed fraction
of female fund managers in the industry could thus be an equilibrium outcome in the sense that the
negative direct flow effect of having a female-managed fund is offset by the positive spill-over effects
on flows in other funds offered by the fund company.
7 Conclusion
This paper examines the conjecture that mutual fund investors exhibit gender bias and prefer to invest
in male-managed funds. Consistent with this conjecture, we find evidence that mutual fund investors
direct significantly less money into female-managed funds. We are able to replicate this finding under
the controlled conditions of a laboratory experiment and can reject several alternative explanations for
lower inflows into female-managed funds. Furthermore, we find that female fund managers follow more
reliable investment styles and we document that performance is identical between male and female fund
managers, while the performance of female managers is more stable than that of male managers. These
results provide no support for the notion that the lower inflows into female-managed funds might be
due to rational statistical discrimination. Rather, our results from an implicit association test suggests
28As this variable might also proxy for the size of the fund company (as it is more likely to have at least one femalemanager if there are simply more other funds), we again include company-level flows as control variable. This capturesthe impact of the size of the fund company (and of all other fund company characteristics) on individual level flows.
27
that there is a gender bias among most of the subjects participating in our experiment. Subjects with
the strongest gender bias (according to the IAT) invest the least in female-managed funds.
Overall, our findings show that gender bias of investors can have a strong impact on financial
markets and help to clarify why female-managed funds receive much lower inflows than male-managed
funds. Furthermore, as managers generating low inflows are not attractive for fund companies to hire,
our results also provide a possible new explanation for the low fraction of female managers in the
mutual fund industry based on customer-driven discrimination.
28
Appendix A: Gender classification
To identify a fund manager’s gender we first extract the manager’s first name from the Morningstar
databases. From a list published by the United States Social Security Administration (SSA) that
contains the most popular first names by gender for the last 10 decades we get 2,179 different male
and 2,515 different female first names that also account for differences in spelling.29 First names that
appear for both sexes are excluded from the SSA-List. We then match this list with the first names
and thereby classify most of the managers as male or female. Remaining names are those we could not
clearly classify as male or female, i.e., foreign names or ambiguous names. We were able to identify
most of the foreign names by asking foreign exchange students from the respective country. For the
remaining cases, we try to identify fund managers’ gender by several internet sources like the fund
prospectus, press releases or photographs that reveal their gender. This leaves us with an identification
rate of 99.39%.
29For further information see http://www.ssa.gov.
29
Appendix B: Brief definitions and data sources of main variables
This table briefly defines the main variables used in the empirical analysis. The data sources are: (i)CRSP: CRSP Survivor-Bias-Free Mutual Fund Database, (ii) CIQ: Capital IQ, (iii) EST: Estimatedor computed by the authors, (iv) EX: Experimental data, (v) KF: Kenneth French Data Library, (vi)LN: LexisNexis, (vii) MSD: Morningstar Direct, (viii) MSP: Morningstar Principia.
Panel A: Measures of fund flows
Variable name Description Source
FundF lowsi,t Computed asTNAi,t−TNAi,t−1·(1+FundReti,t)
TNAi,t−1where TNAi,t denotes fund i’s total
net assets in year t and FundReti,t denotes fund i’s return in year t.
CRSP, EST
AbsF lowi,t Computed as TNAi,t − TNAi,t−1 · (1 +Reti,t). CRSP, EST
ChgMktShri,t Computed asTNAi,t
AggTNAi,t− TNAi,t−1
AggTNAi,t−1where AggTNAi,t denotes the aggregate
assets under management of all funds in the same year and market segment as
fund i.
CRSP, EST
Panel B: Measures of fund performance
Variable name Description Source
FundReti,t A fund’s annual raw net return. CRSP
CAPMi,t Jensen (1968) performance Alpha. We use three years of monthly return data
first to compute factor loadings and then use the last 12 months of realized fund
and factor return data in this period to compute Alphas.
CRSP, KF,
EST
FFi,t Fama and French (1993) performance Alpha. We use three years of monthly return
data first to compute factor loadings and then use the last 12 months of realized
fund and factor return data in this period to compute Alphas.
CRSP, KF,
EST
Cari,t Carhart (1997) performance Alpha. We use three years of monthly return data
first to compute factor loadings and then use the last 12 months of realized fund
and factor return data in this period to compute Alphas.
CRSP, KF,
EST
ShaRi,t Sharpe Ratio computed as a fund’s annual excess return over the risk free rate
divided by the annualized return standard deviation based on monthly return
data.
CRSP, EST
AppRi,t Appraisal Ratio computed as a fund’s four factor abnormal return, Cari,t divided
by the standard deviation of the residuals of the four-factor regression.
CRSP, EST
PerfRanki,t Performance rank of a fund based on its annual return relative to its market
segment in a given year. This variable is normalized to be between zero and one.
The best fund is assigned a rank of one.
CRSP, EST
PerfPersi,m Performance persistence measured as the time series standard deviation of man-
ager m’s performance ranks at fund i. At least three years of performance ranks
are required.
CRSP, EST
Quintile1i,t Piecewise linear regression (PLR) variable, computed as min(PerfRank; 0.2). CRSP, EST
Quintiles2− 4i,t PLR variable, computed as min(PerfRank −Quintile1; 0.8). CRSP, EST
Quintile5i,t PLR variable, computed as min(PerfRank − (Quintile1 +Quintiles2− 4)). CRSP, EST
30
Panel C: Measures of investment behavior
Variable name Description Source
FundRiski,t Fund i’s monthly return standard deviation in year t. CRSP, EST
SysRiski,t Fund i’s factor loading on the market factor from a one factor model in year t. CRSP, EST
UnsysRiski,t Standard deviation of fund i’s residual return from a one factor model in year t. CRSP, EST
TORatioi,t Fund i’s annual turnover ratio in %. CRSP
SVMfi,m Style variability of fund i with respect to a specific factor loading f while manager
m is managing this fund. It is calculated as the standard deviation of a fund’s
yearly factor loadings f over time. Standard deviations are rescaled by the average
factor weighting standard deviation of all funds in the corresponding market
segment over the same period. At least 3 years of consecutive data are required.
CRSP, EST
SVMi Average style variability of fund i calculated as the average of the factor individual
style variability measures, SVMfi,m.
CRSP, EST
Panel D: Main independent variables
Variable name Description Source
Femalei,t Dummy variable equal to one if fund i is managed by a woman in year t, and
zero otherwise.
MSD
FemNewi,t Dummy variable equal to one if a male manager at fund i is replaced by a female
manager in year t, and zero otherwise.
MSD
MgrChgi,t Dummy variable equal to one if there is a manager change at fund i in year t,
and zero otherwise.
MSD
FundSizei,t Logarithm of a fund’s total net assets, ln(TNA+ 1). CRSP, EST
ExpRatioi,t Fund i’s annual expense ratio in %. CRSP
Act12b1i,t Fund i’s actual 12b-1 fees in %. CRSP
MgrTenurei,t Tenure of fund i’s manager in years, computed as difference between year t and
the year in which the manager started managing fund i.
CRSP, EST
FundAgei,t Logarithm of fund i’s age in years (plus one) computed based on the date a fund
was first offered (variable first offer dt).
CRSP, EST
SegmentF lowk,t Average of FundF lowsi,t over all funds i belonging to the same segment k in
year t.
CRSP, EST
CompanyF lowc,t Average of FundF lowsi,t over all funds i belonging to the same fund company c
in year t.
CRSP, EST
MgrAgei,t Logarithm of a fund manager’s age in years (plus one). Data are manually col-
lected from manager biographies.
MSP, MSD,
CIQ
Bachelori,t Dummy variable equal to one if a fund manager has obtained a Bachelor degree,
and zero otherwise. Data are manually collected from manager biographies.
MSP, MSD,
CIQ
MBAi,t Dummy variable equal to one if a fund manager has obtained a Master of Business
Administration (MBA) degree, and zero otherwise. Data are manually collected
from manager biographies.
MSP, MSD,
CIQ
PhDi,t Dummy variable equal to one if a fund manager has obtained a PhD degree, and
zero otherwise. Data are manually collected from manager biographies.
MSP, MSD,
CIQ
ProfQuali,t Dummy variable equal to one if a fund manager has obtained a professional qual-
ification (mainly CFA, but also others such as CFP or CPA), and zero otherwise.
Data are manually collected from manager biographies.
MSP, MSD,
CIQ
LN(1+MedCov)i,t Logarithm of the number of articles on fund i’s manager in year t. Details on the
media data collection process are described in Appendix C.
LN
FemInCompanyc,t Dummy variable equal to one if the fund company c a fund belongs to employs
at least one female fund manager in year t, and zero otherwise.
MSD, CRSP
NumberFemalesc,t Number of female fund manager that the fund company c a fund belongs to
employs in year t.
MSD, CRSP
31
Panel E: Reaction time variables from the experiment
Variable name Description Source
d(R) Difference in mean reaction times in milliseconds between the incompatible and
the compatible configuration in the IAT.
EXP, EST
d(log(R)) Difference in mean log reaction times in milliseconds between the incompatible
and the compatible configuration in the IAT.
EXP, EST
d(S) Difference in mean speed between the compatible and the incompatible configu-
ration. The speed variable is defined as S = 1,000R
.
EXP, EST
Panel F: Other variables from the experiment
Variable name Description Source
FemaleA Dummy variable equal to one if fund A is managed by a female manager, and
zero otherwise.
EXP, EST
FinEconj Dummy variable equal to one if subject j studies finance or economics, and zero
otherwise.
EXP, EST
FinLitj Financial literacy of subject j, computed as the number of right answers that are
given to the 6 financial literacy questions (see Appendix D).
EXP
HighFinLitj Dummy variable equal to one if subject j answered at least 3 out of 6 financial
literacy questions correctly, and zero otherwise.
EXP
SubjIATj Dummy if IAT score of subject j is above the median, and zero otherwise. EXP
SubjGenj Dummy variable equal to one if subject j is female, and zero otherwise. EXP
SubjAgej Subject j’s age in years at the time of the experiment. EXP
SubjMaritalj Dummy variable equal to one if subject j is married, and zero otherwise. EXP
EverInvestj Dummy variable equal to one if subject j ever invested into a mutual fund, and
zero otherwise.
EXP
InstrGenj Dummy variable equal to one if the instructor subject j faced was female, and
zero otherwise.
EXP
32
Appendix C: Media coverage
We use LexisNexis to collect newspaper articles that mention mutual fund managers. We only in-
clude a subset of newspapers in our search strategy to keep the data collection process manageable. We
focus on the top 50 U.S. newspapers according to their print run. Furthermore, we require LexisNexis
to have covered the newspaper since at least the mid 1990s. Additionally, to ensure a regionally bal-
anced panel, we include all regional papers used in Engelberg and Parsons (2011) that are covered in
LexisNexis. The following table shows the newspapers included in our search and the period for which
articles are contained in LexisNexis.
Newspaper Coverage Newspaper Coverage
Atlanta Journal Jan 1991-Dec 2009 Atlanta Constitution Jan 1991-Dec 2009Denver Post Dec 1993-Dec 2009 Houston Chronicle Sep 1991-Dec 2009Las Vegas Review Sep 1996-Dec 2009a Wisconsin State Journal Jan 1992-Dec 2009Minneapolis Star Tribune Sep 1991-Dec 2009 New York Times Jun 1980-Dec 2009Pittsburgh Post-Gazette Mar 1993-Dec 2009 Sacramento Bee Jan 2002-Dec 2009San Antonio Express-News Jan 1996-Dec 2009 San Francisco Chronicle Oct 1989-Dec 2009Seattle Post-Intelligencer Jan 1986-Mar 2009 St. Louis Post-Dispatch Feb 1981-Dec 2009St. Petersburg Time Jan 1987-Dec 2009 Washington D.C. Post Jan 1977-Dec 2009USA Today Jan 1989-Dec 2009 Wall Street Journal May 1973-Dec 2009San Jose Mercury News Jan 1994-Dec 2009 Daily News (New York) Mar 1995-Dec 2009Philadelphia Inquirer Jan 1994-Dec 2009 New York Post Dec 1997-Dec 2009
Dallas Morning News Oct 1992-Dec 2009 Chicago Sun-Times Jan 1992-Dec 2009b
Arkansas Democrat-Gazette Oct 1984-Dec 2009c Augusta Chronicle Jan 1992-Dec 2009d
Austin American-Statesman Jan 1994-Dec 2009 Buffalo News Nov 1992-Dec 2009Christian Science Monitor Jan 1980-Dec 2009 Dayton Daily News Jan 1994-Dec 2009Fresno Bee Jan 1994-Dec 2009 Oklahoman Jan 1992-Dec 2009Palm Beach Post Aug 1988-Dec 2009 Phoenix New Times Jan 1989-Dec 2009Providence Journal-Bulletin Jan 1994-Dec 2009 Record (Bergen County, NJ) Jan 1996-Dec 2009Richmond Times Dispatch Nov 1995-Dec 2009 Salt Lake Tribune Jan 1994-Dec 2009Santa Fe New Mexican Jan 1994-Oct 2011 Tulsa World Dec 1995-Dec 2009Virginian-Pilot Jan 1994-Dec 2009a Stories not available for October 9, 2001.
b Stories not available for November 1992.
c Incomplete coverage for 1992 and 1993.
d Incomplete coverage for June 2000
We search for all articles that mention a manager’s first and last name and require that the first
name appears before the last name with a maximum distance of two letters (to allow for middle initials).
To make sure that we capture fund managers, we only count articles that also contain the word ’equity’
or ’stock’ and ’portfolio’ or ’investment’ or ’fund’. Checking a small sample of the articles that were
identified using this search strategy confirmed that most of them were indeed related to the fund
manager. We do not distinguish between cases in which fund managers were interviewed or are quoted
with their comments and cases in which an article features the success of a fund manager explicitly.
33
Appendix D: Details of the experimental procedure
The experiment took place in 11 individual sessions with a total of 100 students in the McCombs
School of Business Behavioral Laboratory at the University of Texas at Austin. Subjects were recruited
via flyers and announcements made in undergraduate business classes and on Blackboard (a class
management and student communication system used at McCombs). Subjects participated in the
experiment while sitting in front of PC screens that were separated from each other. After all subjects
were seated, a female or male instructor briefly explained the experiment to them. They were told that
the experiment would consist of two parts, a simple investment task (as described in detail in the main
text) and a concentration task (the IAT). Afterwards, a short survey was conducted. Pay consisted of
two parts. The first part was a show-up fee of 4 USD, the second part was a payoff that depended
on the return of their investment decision in one randomly drawn round. The return was determined
based on the actual annual return from CRSP of the funds they could choose from in that specific
round. One experimental unit in the investment task was equivalent to 5.50 USD. Subjects earned on
average 24 USD, with a maximum (minium) of 38 (4) USD.
The concentration task consists of an IAT which we designed to uncover prejudice against women
in finance. Following Greenwald, McGhee, and Schwartz (1998), the IAT is played in seven rounds and
two versions. Out of the seven rounds, two rounds are test rounds that are evaluated, while the other
five rounds are practice rounds. First, two practice rounds with 20 trials each are played to familiarize
subjects with the tasks. In the first (second) round, only items belonging to the categories ’female’ and
’male’ (’marketing’ and ’finance’) have to be sorted (see Table 8). Then, another practice round with
20 trials was administered in which subjects are asked to categorize items in a combined task, i.e., to
categorize items into the ’male/female’ and ’marketing/finance’ categories. After these three practice
rounds, a test round with 40 trials which was otherwise identical to the third practice round is played.
Then, two more practice rounds 5 and 6 with 20 trials each follow that are similar to the test rounds 1
and 3. However, one of the categories is exchanged from the left to the right side of the screen. Finally,
round 7 is another test round with 40 trials, which is identical to the last practice round. Our main
results in the paper are based on the reaction times subjects achieve in the two test rounds 4 and 7.
Results are very similar if we also include results from the two practice rounds 3 and 6.
The final survey consisted of questions on subjects’ demographic characteristics, a question whether
they had any investment experience, and a short financial literacy test. This test consists of financial
literacy questions that are also used in van Rooij, Lusardi, and Alessie (2011).
34
References
Adams, Renee B., and Patricia Funk, 2012, Beyond the glass ceiling: Does gender matter?,ManagementScience 58, 219–235.
Atkinson, Stanely M., Samantha Boyce Baird, and Melissa B. Frye, 2003, Do female mutual fundmanagers manage differently?, Journal of Financial Research 26, 1–18.
Baer, Michaela, Alexander Kempf, and Stefan Ruenzi, 2011, Is a team different from the sum of itsparts? team management in the mutual fund industry, Review of Finance 15, 359–396.
Bailey, Warren, Alok Kumar, and David Ng, 2011, Behavioral biases of mutual fund investors, Journalof Financial Economics 102, 1–27.
Baks, Klaas P., 2003, On the performance of mutual fund managers, Working Paper.
Banaji, Mahzarin R., and Anthony G. Greenwald, 1995, Implicit gender stereotyping in judgements offame, Journal of Personality and Social Psychology 68, 181–198.
Barber, Brad M., and Terrance Odean, 2001, Boys will be boys: Gender, overconfidence, and commonstock investment, Quarterly Journal of Economics 116, 261–292.
, and Lu Zheng, 2005, Out of sight, out of mind: The effects of expenses on mutual fund flows,Journal of Business 78, 2095–2120.
Beaman, Lori, Rahabendra Chattopadhyay, Esther Duflo, Rohini Pande, and Petia Topalova, 2009,Powerful women: Does exposure reduce bias?, Quarterly Journal of Economics 124, 1497–1540.
Becker, Gary S., 1971, The Economics of Discrimination (University of Chicago Press: Chicago) 2ndedn.
Berk, Jonathan B., and Richard C. Green, 2004, Mutual fund flows and performance in rational markets,Journal of Political Economy 112, 1269–1295.
Bertrand, Marianne, Dolly Chugh, and Sendhil Mullainathan, 2005, Implicit discrimination, AmericanEconomic Review 95, 94–98.
Bertrand, Marianne, Claudia Goldin, and Lawrence F. Katz, 2010, Dynamics of the gender gap foryoung professionals in the financial and corporate sectors, American Economic Journal: AppliedEconomics 2, 228–255.
Bertrand, Marianne, and Kevin F. Hallock, 2001, The gender gap in top corporate jobs, Industrial andLabor Relations Review 55, 3–21.
Betrand, Marianne, and Antoinette Schoar, 2003, Managing with style: The effect of managers on firmpolicies, Quarterly Journal of Economics 118, 1169–1208.
Blair, Irene V., and Mahzarin R. Banaji, 1996, Implicit gender stereotyping in judgments of fame,Journal of Personality and Social Psychology 70, 1142–1163.
Brown, Keith C., W. Harlow, and Hanjiang Zhang, 2014, Investment style volatility and mutual fundperformance, Working Paper.
35
Byrnes, James P., David C. Miller, and William D. Schafer, 1999, Gender differences in risk taking: Ameta-analysis, Psychological Bulletin 125, 367–383.
Carhart, M. M., 1997, On persistence in mutual fund performance, Journal of Finance 52, 57–82.
Carney, Dana R., Brian A. Nosek, Anthony G. Greenwald, and Mahzarin R. Banaji, 2007, Implicitassociation test (IAT), Working Paper.
Chen, Joseph, Harrison Hong, Ming Huang, and Jeffrey D. Kubik, 2004, Does fund size erode perfor-mance? Liquidity, organizational diseconomies and active money management, American EconomicReview 94, 1276–1302.
Chevalier, Judith, and Glenn Ellison, 1999, Career concerns of mutual fund managers, Quarterly Jour-nal of Economics 114, 389–432.
Choi, James J., David Laibson, and Brigitte C. Madrian, 2011, Why does the law of one price fail? Anexperiment on index mutual funds, Review of Financial Studies 23, 1405–1432.
Christoffersen, Susan, Richard Evans, and David Musto, 2013, What do consumers’ fund flows maxi-mize? Evidence from their brokers’ incentives, Journal of Finance 68, 201–235.
Cronqvist, Henrik, 2006, Advertising and portfolio choice, Working Paper.
Croson, Rachel, and Uri Gneezy, 2009, Gender differences in preferences, Journal of Economic Litera-ture 47, 1–27.
Cunningham, William A., Kristopher J. Preacher, and Mahzarin R. Banaji, 2001, Implicit attitudemeasures: Consistency, stability, and convergent validity, Psychological Science 12, 163–170.
Daniel, Kent, Mark Grinblatt, Sheridan Titman, and Russ Wermers, 1997, Measuring mutual fundperformance with characteristic-based benchmarks, Journal of Finance 52, 1035–1058.
Engelberg, Joseph E., and Christopher Parsons, 2011, The causal impact of media in financial markets,Journal of Finance 66, 67–97.
Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the return on bonds andstocks, Journal of Financial Economics 33, 3–53.
Fama, Eugene F., and James MacBeth, 1973, Risk, return and equilibrium: Empirical tests, Journal ofPolitical Economy 81, 607–636.
Folkard, Simon, 1976, Time of day and level of processing, Memory & Cognition 7, 247–252.
Gallaher, Steven, Ron Kaniel, and Laura T. Starks, 2015, Advertising and mutual funds: From familiesto individual funds, CEPR Discussion Paper No. 10329.
Glode, Vincent, 2011, Why mutual funds ’underperform’, Journal of Financial Economics 99, 546–559.
Goldin, Claudia, and Cecilia Rouse, 2000, Orchestrating impartiality: The impact of ’blind’ auditionson female musicians, American Economic Review 90, 715–741.
Graham, John R., Campbell R. Harvey, and Manju Puri, 2013, Managerial attitudes and corporateactions, Journal of Financial Economics 109, 103–121.
36
Green, Clifton, Narasimhan Jegadeesh, and Yue Tang, 2009, Gender and job performance: Evidencefrom wall street, Financial Analysts Journal 65, 65–78.
Greenwald, Anthony G., Mahzarin R. Banaji, L. A. Rudman, S. D. Farnham, B. A. Nosek, and D. S.Mellott, 2002, A unified theory of implicit attitudes, stereotypes, self-esteem, and self-concept, Psy-chological Review 109, 3–25.
Greenwald, Anthony G., Debbie E. McGhee, and Jordan L. K. Schwartz, 1998, Measuring individualdifferences in implicit cognition: The implicit association test, Journal of Personality and SocialPsychology 74, 1464–1480.
Greenwald, Anthony G., T. Andrew Poehlman, Eric Luis Uhlmann, and Mahzarin R. Banaji, 2009, Un-derstanding and using the implicit association test: III. Meta-analysis of predictive validity, Journalof Personality and Social Psychology 97, 17–41.
Holzer, Harry J., and Keith R. Ihlanfeldt, 1998, Customer discrimination and employment outcomesfor minority workers, Quarterly Journal of Economics 113, 835–867.
Jain, Prem C., and Joanna Shuang Wu, 2000, Truth in mutual fund advertising: Evidence on futureperformance and fund flows, Journal of Finance 55, 937–958.
Jensen, Michael C., 1968, The performance of mutual funds in the period 1955-1964, Journal of Finance23, 389–416.
Kaniel, Ron, Laura Starks, and Vasudha Vasudevan, 2007, Headlines and bottom lines: Attention andlearning effects from media coverage and mutual funds, Working Paper.
Kumar, Alok, 2010, Self-selection and the forecasting abilities of female equity analysts, Journal ofAccounting Research 48, 393–436.
, Alexandra Niessen-Ruenzi, and Oliver Spalt, 2015, What is in a name? Mutual fund flowswhen managers have foreign-sounding names, Review of Financial Studies forthcoming.
Lane, Kristin A., Mahzarin A. Banaji, Brian A. Nosek, and Anthony G. Greenwald, 2007, Understand-ing and using the implicit association test: IV What we know (so far) about the method, in ImplicitMeasures of Attitudes (Taylor & Francis Lea).
Massa, Massimo, Jonathan Reuter, and Eric Zitzewitz, 2010, When should firms share credit withemployees? Evidence from anonymously managed mutual funds, Journal of Financial Economics95, 400–424.
Nardinelli, Clark, and Curtis Simon, 1990, Customer racial discrimination in the market for memora-bilia: The case of baseball, Quarterly Journal of Economics 105, 575–595.
National Council for Research on Women, 2009, Women in fund management, Report, New York.
Neumark, David, 1996, Sex discrimination in restaurant hiring: an audit study, Quarterly Journal ofEconomics 111, 915–942.
Newton, David, and Mikhail Simutin, 2014, Of age, sex, and money: Insights from corporate officercompensation on the wage inequality between genders, Management Science forthcoming.
Niederle, Muriel, and Lise Vesterlund, 2007, Do women shy away from competition?, Quarterly Journalof Economics 122, 1067–1101.
37
Patel, Jay, Richard Zeckhauser, and Darryl Hendricks, 1991, The rationality struggle: Illustrations fromfinancial markets, American Economic Review 81, 232–236.
Paulus, Paul B., Angela B. Annis, John J. Seta, Janette K. Schkade, and Robert W. Matthews, 1976,Density does affect task performance, Journal of Personality and Social Psychology 34, 248–253.
Phelps, Edmund, 1972, The statistical theory of racism and sexism, American Economic Review 62,659–661.
Polachek, Solomon William, 1981, Occupational self-selection: A human capital approach to sex differ-ences in occupational structure, Review of Economics and Statistics 63, 60–69.
Reuter, Jonathan, and Eric Zitzewitz, 2013, How much does size erode mutual fund performance? Aregression discontinuity approach, Working Paper.
Sharpe, William F., 1966, Mutual fund performance, Journal of Business 39, 119–138.
Sialm, Clemens, and T. Mandy Tham, 2015, Spillover effects in mutual fund companies, ManagementScience forthcoming.
Sirri, Erik R., and Peter Tufano, 1998, Costly search and mutual fund flows, Journal of Finance 53,1589–1622.
Spiegel, Matthew I., and Hong Zhang, 2013, Mutual fund risk and market share adjusted fund flows,Journal of Financial Economics 108, 506–528.
Sutter, Matthias, and Daniela Gatzle-Rutzler, 2014, Gender differences in the willingness to competeemerge early in life and persist, Management Science forthcoming.
Tajfel, H., 1970, Experiments in intergroup discrimination, Scientific American 223, 96–102.
Treynor, J.L., and F. Black, 1973, How to use security analysis to improve portfolio selection, Journalof Business 46, 66–86.
van Rooij, Maarten, Annamaria Lusardi, and Rob Alessie, 2011, Financial literacy and stock marketparticipation, Journal of Financial Economics 101, 449–471.
Wang, P., 1994, Brokers still treat men better than women, Money 23, 108–110.
Wolfers, Justin, 2006, Diagnosing discrimination: Stock returns and CEO gender, Journal of the Eu-ropean Economic Association 4, 531–541.
38
Table 1: Descriptive Statistics
Panel A Mean Median SD p1 p99 Obs.
(1) (2) (3) (4) (5) (6)
Femalei,t 0.108 0.000 0.310 0.000 1.000 13302
FundF lowsi,t (in percent) 0.280 0.052 1.094 -0.561 1.441 13302
AbsF lowi,t 64.653 2.774 258.085 -411.027 1330.419 12974
ChgMktShri,t -0.001 0.000 6.621 -8.362 8.462 13302
FundReturni,t 0.040 0.047 0.278 -0.524 0.720 13302
CAPMi,t -0.063 -0.074 1.165 -3.370 3.316 13278
FFi,t -0.134 -0.096 1.134 -3.690 2.989 13278
Cari,t -0.157 -0.103 1.177 -4.091 2.938 13278
ShaRi,t 0.184 0.041 1.398 -2.199 3.957 12916
AppRi,t -0.001 -0.000 0.008 -0.027 0.017 13278
FundSizei,t (in Millions) 980.8 172.1 2987 1.251 13565 13302
ExpRatioi,t (in percent) 0.014 0.013 0.014 0.002 0.036 13291
Act12b1i,t (in percent) 0.003 0.003 0.003 0.000 0.010 8090
TORatioi,t 1.009 0.661 1.626 0.030 6.520 13243
FundRiski,t 0.050 0.044 0.027 0.014 0.145 13296
SysRiski,t 0.994 0.949 0.417 0.179 2.435 13278
UnsysRiski,t 6.180 2.458 15.464 0.093 53.726 13278
SVMi 1.000 0.851 0.613 0.237 3.519 2272
FundAgei,t(in years) 13.102 9.000 12.622 3.000 68.000 13302
MgrAgei,t (in years) 45.658 45.000 8.703 28.000 68.000 10630
MgrTenurei,t (in years) 5.863 5.000 4.617 0.000 13.000 13298
Bachelori,t 0.998 1.000 0.039 1.000 1.000 10630
MBAi,t 0.556 1.000 0.497 0.000 1.000 10630
PhDi,t 0.056 0.000 0.231 0.000 1.000 10630
ProfQuali,t 0.521 1.000 0.500 0.000 1.000 10630
MedCovi,t 2.021 0.000 7.306 0.000 33.000 13302
39
Table 1: continued
Panel B Female Manager Male Manager Difference
(1) (2) (3)
FundF lowsi,t 0.19 0.29 −0.10∗∗∗
FundReturni,t 0.05 0.06 0.01
CAPMi,t −0.09 0.05 −0.04
FFi,t −0.06 −0.06 0.00
Cari,t −0.06 −0.07 0.01
ShaRi,t 0.27 0.20 0.07∗
AppRi,t −0.00 −0.00 0.00
FundSizei,t 573.07 711.01 −137.94∗∗∗
ExpRatioi,t 1.46 1.44 0.02
Act12b1i,t 0.32 0.28 0.04∗∗∗
TORatioi,t 0.95 1.07 −0.12∗∗
FundRiski,t 0.05 0.05 0.00
SysRiski,t 0.98 0.99 −0.01
UnsysRiski,t 6.31 6.27 0.04
FundAgei,t 10.89 10.33 0.55∗∗
MgrAgei,t 43.06 45.28 −2.22∗∗∗
MgrTenurei,t 4.90 5.99 −1.09∗∗∗
Bachelori,t 99.59 99.90 −0.31∗∗
MBAi,t 56.08 55.04 1.04
PhDi,t 1.78 6.53 −4.75∗∗∗
ProfQuali,t 0.53 0.53 0.00
MedCovi,t 0.96 2.15 −1.19∗∗∗
Notes: Panel A of this table shows fund characteristics based on our sample of all single-managed U.S. equity funds from January
1993 to December 2009. Means, medians, standard deviations (SD), bottom percentile (p1), upper percentile (p99), and the number
of observations (Obs.) are reported. The detailed description of the variables listed in the first column is contained in Appendix B.
Panel B of this table shows average characteristics for female-managed funds, average characteristics for male-managed funds, and
the difference between the average characteristics of female and male fund managers. Significance is calculated based on a two-sided
t-test. ∗∗∗ 1% significance, ∗∗ 5% significance, ∗ 10% significance.
40
Table
2:FundFlows
NLD
WLD
RankRet
RankCar
USE
FMB
YC
FYC
NLS
Perf.Interactions
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
Fem
ale
i,t
–0.120
–0.121
–0.111
–0.104
–0.116
–0.106
–0.111
–0.107
–0.096
–0.123
–0.163
(–4.32)
(–4.23)
(–3.95)
(–3.74)
(–3.37)
(–4.03)
(–4.18)
(–3.47)
(–3.58)
(–4.32)
(–2.64)
FundFlows i
,t−1
0.046
0.032
0.042
0.024
0.064
0.032
0.043
0.055
0.046
0.050
(5.12)
(3.67)
(4.79)
(2.02)
(3.72)
(3.69)
(3.46)
(6.43)
(5.12)
(6.04)
FundRet
i,t−
10.329
0.294
0.324
(4.47)
(3.37)
(4.33)
Per
fRanki,t−
1–0.243
–0.061
–0.266
–0.218
–0.243
–0.249
–0.465
–1.056
(–1.58)
(–0.40)
(–1.42)
(–1.47)
(–1.35)
(–1.74)
(–3.45)
(–6.33)
Per
fRank2 i,t−
10.812
0.559
0.938
0.740
0.812
0.809
1.139
1.754
(5.07)
(3.40)
(4.63)
(4.01)
(3.73)
(4.45)
(7.44)
(9.10)
FundSize i
,t−1
–0.134
–0.139
–0.144
–0.141
–0.146
–0.158
–0.144
–0.158
–0.497
–0.139
–0.121
(–10.07)
(–10.09)
(–10.48)
(–10.33)
(–8.40)
(–12.51)
(–11.40)
(–12.90)
(–3.48)
(–10.10)
(–9.25)
TORatioi,t−
10.059
0.058
0.061
0.056
0.091
0.065
0.061
0.059
0.054
0.058
0.058
(3.61)
(3.65)
(3.72)
(3.44)
(2.81)
(4.36)
(8.50)
(5.93)
(3.60)
(3.65)
(3.70)
FundRiski,t−
1–0.571
–0.455
0.291
–0.938
0.306
1.443
0.291
0.598
–0.553
–0.453
–0.480
(–0.87)
(–0.67)
(0.43)
(–1.40)
(0.34)
(1.31)
(0.48)
(0.81)
(–0.84)
(–0.66)
(–0.72)
ExpRatioi,t−
13.980
4.621
3.987
4.887
4.367
–3.897
3.987
0.863
4.863
4.603
4.342
(1.21)
(1.45)
(1.23)
(1.47)
(1.21)
(–1.53)
(0.91)
(0.23
(1.56)
(1.45)
(1.35)
FundAge i
,t−1
–0.067
–0.025
–0.002
–0.022
0.013
–0.014
–0.002
–0.014
–0.035
–0.025
–0.021
(–3.74)
(–1.29)
(–0.13)
(–1.19)
(0.68)
(–0.74)
(–0.12)
(–0.69)
(–1.99)
(–1.28)
(–1.13)
Seg
men
tFlow
k,t
0.152
0.128
0.138
0.139
–0.071
0.065
0.138
0.129
0.141
0.128
0.144
(3.20)
(2.85)
(3.13)
(3.09)
(–1.47)
(0.23)
(1.60)
(1.33)
(3.20)
(2.85)
(3.27)
CompanyFlow
c,t
0.002
0.000
0.000
0.000
0.001
0.022
0.000
0.004
–0.000
0.000
0.000
(1.03)
(0.03)
(0.08)
(0.14)
(0.51)
(3.04)
(0.07)
(2.12)
(–0.02)
(0.03)
(0.29)
FundSize2 i,
t−1
0.049
(1.96)
FundSize3 i,
t−1
–0.002
(–1.11)
FundRet
·Fem
ale
i,t−
10.083
(0.98)
Per
fRank·F
emale
i,t−
10.790
(2.25)
Per
fRank2·F
emale
i,t−
1–1.043
(–2.59)
(adj./avg.)
R2
0.146
0.157
0.176
0.169
0.204
0.143
0.176
0.095
0.197
0.146
0.194
Observations
13265
12301
12301
12232
8223
12334
12301
12334
12279
13265
12301
41
Table 2: continued
Notes: This table shows the estimates of percentage fund flows, FundF lowsi,t, regressed on a female fund manager
dummy, as well as fund and segment characteristics. Fund flows are calculated by subtracting the internal growth of a
fund due to the returns earned on assets under management from the total growth rate of the fund’s total net-assets under
management. Femalei,t is a dummy variable that takes on the value one, if a fund i is managed by a female manager in
year t, and zero otherwise. FundReti,t−1 denotes fund i’s lagged net return. FundSizei,t−1 is the lagged natural logarithm
of the fund’s size in million USD and TORatioi,t−1 is the fund’s lagged turnover rate. FundRiski,t−1 is the lagged return
time series standard deviation of fund i. ExpRatioi,t−1 is the fund’s lagged total expense ratio. FundAgei,t−1 is the lagged
natural logarithm of fund i’s age in years. SegmentF lowk,t is the average growth rate of all funds in fund i’s market
segment k due to flows in year t. CompanyF lowc,t is the average growth rate of all funds in fund i’s fund company c due
to flows in year t. SegmentF lowk,t and CompanyF lowc,t are calculated net of the flows into fund i. Column (1) reports
results without the lagged dependent variable (NLD), while Column (2) presents results including the lagged dependent
variable (WLD). In Columns (3) to (9) and (11), we include the performance rank of fund i in the previous year t − 1,
PerfRanki,t−1, as well as the squared performance rank of fund i in the previous year t− 1, PerfRank2i,t−1 relative to
all other funds in the same market segment to capture the non-linearity of the performance-flow relationship. In Columns
(3) and (4), performance ranks are computed based on raw returns (RankRet) or based on Carhart (1997) four factor
Alphas (RankCar), respectively. Results in Column (5) are obtained from a subsample of funds investing in U.S. equities
(USE) only. Results in Column (6) are based on Fama and MacBeth (1973) regressions (FMB). In Columns (7) and (8),
standard errors are clustered at the year level (YC) and at the fund and year level (FYC), respectively. In Column (9),
we include fund size to the power of two and three to capture a non-linear impact of size (NLS). In Columns (10) and
(11), we interact the female dummy variable with lagged performance. Regressions are estimated with time (except in
Column (6)), segment and fund company fixed effects. t-statistics are in parentheses. In Columns (1) to (5) and (9) to
(11), standard errors are clustered at the fund level.
42
Table 3: Fund Flows: Alternative Explanations and Robustness
Panel A: Alternative Explanations
Manager Manager Media Adver- Broker Retail Instl.Change Char. Coverage tising Channel Fund Fund
(1) (2) (3) (4) (5) (6) (7)FemNewi,t−1 –0.127
(–1.93)MgrChgi,t−1 –0.011
(–0.40)Femalei,t –0.119 –0.108 –0.120 –0.120 –0.155 –0.138
(–3.99) (–3.91) (–3.35) (–3.68) (–3.84) (–1.34)MBAi,t 0.001
(0.04)PhDi,t –0.056
(–1.59)ProfQuali,t 0.014
(0.50)MgrAgei,t –0.003
(–1.50)MgrTenurei,t−1 0.011
(3.86)LN(1 +MedCov)i,t−1 0.046
(3.04)NoLoad · Femi,t 0.024
(0.47)NoLoadi,t 0.028
(1.04)Act12b1i,t –16.210
(–1.90)Controls yes yes yes yes yes yes yesadj./Pseudo R2 0.193 0.169 0.194 0.236 0.194 0.187 0.445Observations 12300 9787 12301 7503 12299 6973 1484
Panel B: Robustness
Alternative Flow Measures PLRet PLCar Year Year Good Bad≤ 2001 >2001 Market Market
(1) (2) (3) (4) (5) (6) (7) (8) (9)Femalei,t –14.270 –0.009 –0.004 –0.112 –0.108 –0.085 –0.201 –0.124 –0.097
(–1.99) (–3.02) (–4.33) (–4.00) (–3.87) (–2.09) (–4.72) (–3.62) (–2.20)Quintile1i,t−1 0.193 0.748
(0.68) (3.17)Quintile2− 4i,t−1 0.381 0.229
(7.53) (4.56)Quintile5i,t−1 2.373 2.474
(6.65) (5.72)Controls yes yes yes yes yes yes yes yes yesadj./Pseudo R2 0.299 0.007 0.123 0.178 0.172 0.104 0.238 0.177 0.339Observations 11890 15376 247630 12301 12232 6614 5687 8759 3542
Panel C: Propensity Score Matching Analysis
Nearest Radius Kernel Strati-Neighbor fication
(1) (2) (3) (4)Femalei,t –0.070 –0.051 –0.094 –0.115
(–2.04) (–2.32) (–4.71) (–4.79)Number of matches 1332 1332 1332 1226
43
Table 3: continued
Notes: In this table, we use the same baseline specification as in Column (3) of Table 2. In Column (1) of Panel A,
we replace our female indicator variable, Femalei,t, with a variable that is equal to one if a male manager at fund i is
replaced by a female manager in year t− 1, FemNewi,t−1, and zero otherwise. MgrChgi,t−1 is a dummy variable equal
to one if a manager change occurred at fund i in year t− 1. In Column (3), we add the logarithm of a manager’s media
coverage, LN(1+MedCov)i,t−1, as a control variable. In Column (4), we add 12b1 fees (Act12b1i,t) as a control variable.
In Column (5), we interact our female indicator variable with a dummy variable equal to one, if a fund charges no load
fees, NoLoadi,t, and zero otherwise. In Columns (6) and (7), we restrict our sample to funds that are declared as retail
(institutional) funds, respectively. In Panel B, we use absolute fund flows, AbsF lowsi,t, (Column (1)), the change of a
fund’s market share, ChgMktShri,t, (Column (2)), both as defined in Appendix B, and monthly instead of yearly fund
flows (Column (3)) as alterative dependent variables. In Columns (4) and (5) we capture the nonlinear performance flow
relationship by a piecewise linear regression approach instead of squared performance ranks. Ranks are based on returns
(PLRet) and Carhart (1997) four factor Alphas (PLCar), respectively. Results in the last four columns of Panel B are
based on subsamples of funds till 2001 (Column (6)), after 2001 (Column (7)), in years following positive market returns
(Column (8)) and in years following negative market returns (Column (9)), respectively. Panel C reports results from
a propensity score matching analysis where we match based on segment, size, and past fund returns. t-statistics are in
parentheses.
44
Table 4: Gender Differences in Investment Behavior
Panel A: Risk Taking and Trading Activity
FundRiski,t SysRiski,t UnsysRiski,t TORatioi,t
(1) (2) (3) (4)
Femalei,t –0.000 –0.004 –0.424 –0.020
(–0.44) (–0.31) (–1.16) (–0.62)
FundSizei,t−1 0.001 0.023 –0.188 –0.078
(5.35) (5.30) (–1.24) (–6.31)
ExpRatioi,t−1 0.072 1.163 96.950 –0.004
(1.85) (2.59) (1.36) (–0.00)
FundAgei,t−1 –0.001 –0.015 0.082 0.030
(–3.20) (–1.87) (0.23) (1.29)
FundReti,t−1 0.009 0.163 3.891 0.113
(5.65) (6.54) (2.58) (1.03)
MgrTenurei,t−1 –0.000 –0.006 0.026 –0.019
(–4.75) (–5.24) (0.59) (–4.42)
adj. R2 0.609 0.334 0.319 0.490
Observations 15153 15122 15122 15048
Panel B: Style Variability
SVMi SVMSMBi SVMHML
i SVMMOMi
Female Manager 0.8748 0.8789 0.8750 0.8706
Male Manager 1.0059 1.0057 1.0059 1.0061
Difference −0.1311∗∗∗ −0.1268∗∗∗ −0.1309∗∗∗ −0.1355∗∗∗
Notes: In Panel A of this table, the dependent variable is one of the following: the fund’s total risk measured by its return time series
standard deviation, FundRiski,t, the fund’s systematic risk, SysRiski,t, defined as the factor loading on the market factor from the
Jensen (1968) one-factor model, the fund’s unsystematic risk, UnsysRiski,t, defined as the standard deviation of the residuals from
the Jensen (1968) one-factor model, and the fund’s turnover ratio, TORatioi,t. Femalei,t is a dummy variable that takes on the
value one, if fund i is managed by a female manager in year t, and zero otherwise. FundSizei,t−1 is the lagged natural logarithm of
the fund’s size in million USD. ExpRatioi,t−1 is a fund’s lagged total expense ratio. FundAgei,t−1 is the lagged natural logarithm
of fund i’s age in years. FundReti,t−1 is a fund’s lagged raw return. MgrTenurei,t is the fund manager’s tenure with the fund in
years. The regressions are estimated with time, segment, and fund company fixed effects. t-statistics are in parentheses. Standard
errors are clustered at the fund level. Panel B shows the average style variability of female and male-managed funds for the aggregate
style variability measure (Column 1) as well as for the factor individual style variability measures (Columns 2 to 4). The factor
individual style variability measures are defined as the rescaled time series standard deviations of a fund’s factor loading on the
SMB, the HML, and the momentum factor from the Carhart (1997) four-factor model. The aggregate style variability measure is
defined as the average of the three factor individual style variability measures. Differences in style variability between female and
male fund managers are given in the third line. Significance is calculated based on a two-sided t-test. ∗∗∗ 1% significance, ∗∗ 5%
significance, ∗ 10% significance.
45
Table 5: Gender and Fund Performance
Panel A: Fund Performance - Multivariate Evidence
FundReti,t CAPMi,t FFi,t Cari,t ShaRi,t AppRi,t
(1) (2) (3) (4) (5) (6)
Femalei,t –0.003 –0.006 –0.001 –0.001 –0.005 –0.000
(–0.80) (–0.92) (–0.20) (–0.18) (–0.18) (–0.46)
FundSizei,t−1 –0.013 –0.011 –0.005 –0.006 –0.068 –0.000
(–12.65) (–7.80) (–3.65) (–4.51) (–10.89) (–2.83)
ExpRatioi,t−1 –0.329 –0.579 –0.466 –0.396 –1.065 0.010
(–1.56) (–2.53) (–1.55) (–0.87) (–0.87) (1.16)
FundAgei,t−1 0.002 0.001 –0.007 –0.007 –0.016 –0.000
(1.20) (0.42) (–2.68) (–2.42) (–1.33) (–0.69)
MgrTenurei,t−1 0.001 0.000 –0.000 0.000 0.009 0.000
(3.68) (0.69) (–0.08) (0.96) (3.61) (1.41)
R2 0.611 0.167 0.154 0.163 0.606 0.004
Observations 16509 9804 9804 9803 16116 18181
Panel B: Robustness
FundReti,t CAPMi,t FFi,t Cari,t ShaRi,t AppRi,t
Femalei,t (1) (2) (3) (4) (5) (6)
B.1 Fund Chars. –0.002 –0.008 –0.001 –0.002 0.006 –0.000
(–0.41) (–1.10) (–0.18) (–0.27) (0.17) (–0.99)
R2 0.621 0.178 0.163 0.176 0.624 0.050
Observations 12483 9000 9000 8999 12165 13765
B.2 Manager Chars. –0.004 –0.006 0.001 0.001 –0.020 –0.000
(–0.82) (–0.77) (0.16) (0.10) (–0.63) (–0.41)
R2 0.630 0.171 0.159 0.168 0.622 0.006
Observations 12990 7811 7811 7810 12677 14348
B.3 YC –0.003 –0.006 –0.001 –0.001 –0.005 –0.000
(–0.61) (–0.98) (–0.20) (–0.20) (–0.13) (–0.47)
R2 0.611 0.167 0.154 0.163 0.606 0.004
Observations 16509 9804 9804 9803 16116 18181
B.4 FYC –0.000 –0.002 0.001 –0.001 0.028 –0.000
(–0.47) (–0.32) (0.16) (–0.15) (0.73) (–0.57)
R2 0.004 0.109 0.077 0.084 0.591 0.008
Observations 18181 9822 9822 9821 16156 18229
B.5 FMB 0.001 –0.000 0.002 0.000 0.004 –0.000
(0.17) (–0.06) (0.56) (0.12) (0.15) (–0.14)
R2 0.223 0.185 0.165 0.164 0.210 0.074
Observations 16549 9822 9822 9821 16156 18229
46
Table 5: continued
Panel C: Fund Performance - Portfolio Evidence
Equal-Weighted Value-Weighted
CAPMf−mt FF f−m
t Carf−mt CAPMf−m
t FF f−mt Carf−m
t
(1) (2) (3) (4) (5) (6)
Alphat 0.000 0.000 0.000 –0.001 –0.000 –0.001
(0.05) (0.77) (0.09) (–1.61) (–0.70) (–1.20)
MKTRFt 0.019 0.010 0.019 0.035 0.017 0.028
(3.40) (1.77) (3.23) (3.32) (1.63) (2.62)
SMBt 0.011 0.009 0.003 0.000
(1.60) (1.33) (0.26) (0.03)
HMLt –0.034 –0.028 –0.084 –0.075
(–4.62) (–3.77) (–6.13) (–5.45)
MOMt 0.019 0.025
(4.16) (2.94)
R2 0.047 0.165 0.225 0.045 0.200 0.227
Observations 216 216 216 216 216 216
Panel D: Performance Persistence
Female Male Difference
FundReti,t 0.2274 0.2452 −0.0178 (−1.65)
CAPMi,t 0.2565 0.2700 −0.0135 (−1.98)
FFi,t 0.2542 0.2712 −0.0170 (−1.97)
Cari,t 0.2410 0.2637 −0.0227 (−2.46)
ShaRi,t 0.2517 0.2524 −0.0007 (−0.95)
AppRi,t 0.2360 0.2591 −0.0231 (−2.11)
Notes: In Panel A of this table, the performance of a fund computed as the raw return (FundReti,t), the Jensen (1968) Alpha
(CAPMi,t), the Fama and French (1993) three-factor Alpha (FFi,t), the Carhart (1997) four-factor Alpha (Cari,t), the Sharpe
(1966) Ratio (ShaRi,t), or a modified version of the Treynor and Black (1973) Appraisal Ratio (AppRi,t), all as defined in Appendix
B, is the dependent variable. Femalei,t is a dummy variable that takes on the value one, if a fund i is managed by a female manager
in year t, and zero otherwise. MgrTenurei,t is the fund manager’s tenure with the fund in years. All other controls are defined as
in the previous tables. Panel B presents the coefficient and t-statistic on Femalei,t in regressions including the same controls as in
Panel A from various robustness checks. In B.1, we add TORatioi,t−1, FundF lowsi,t−1, FundRiski,t−1, and the lagged dependent
variable as controls. In B.2, we add MgrAgei,t, MBAi,t, PhDi,t, and ProfQuali,t as controls. Results in B.3 (B.4) are obtained by
clustering standard errors at the year level (YC) and the year and fund level (FYC). In B.5, results are obtained by estimating Fama
and MacBeth (1973) regressions. Panel C shows results from a regression with the equal weighted and value weighted return of a
difference portfolio that is long in all female-managed funds and short in all male-managed funds as dependent variable. Difference
returns are regressed on the market factor, MKTRFt, the size factor, SMBt, the value factor, HMLt, and the momentum factor,
MOMt. Panel D contains the average time series standard deviation over performance ranks of female- and male-managed funds for
various performance measures and the difference between female and male fund managers. t-statistics are in parentheses. Regressions
are estimated with time, segment, and fund company fixed effects. Standard errors are clustered at the fund level.
47
Table 6: Subject Characteristics
Panel A: Main Field of Study Number Percentage
Accounting 13 13.00%
Economics 5 5.00%
Finance 43 43.00%
Management Information Systems 9 9.00%
Marketing 10 10.00%
Other 20 20.00%
Panel B: Age in Years Number Percentage
18 to 19 8 8.00%
20 30 30.00%
21 30 30.00%
22 21 21.00%
> 23 12 12.00%
Panel C: Marital Status Number Percentage
Single 97 97.00%
Married/Engaged 3 3.00%
Panel D: Gender Number Percentage
Female 49 49.00%
Male 51 51.00%
Notes: This table shows summary statistics of subjects’ characteristics in our experiment. Panel A displays the number and
percentage of subjects with different main fields of study. The “Other” category mainly includes students in “International
Business” or “Supply Chain Management” as well as students from non-business fields like “Geography”, “Literature”, or
“Physical Therapy”. Panel B contains the number and percentage of subjects in different age brackets. Panel C provides
number and percentage of subjects depending on their marital status and Panel D contains number and percentage of
subjects that belong to each gender category.
48
Table 7: Investment Decisions
Female Manager Male Manager Difference (F-M) Obs.
% invested into fund A
(1) (2) (3) (4)
Panel A: All subjects 41.43 48.85 –7.42∗∗∗ 484
Panel B: Gender
Males 35.77 46.23 –10.47∗∗∗ 252
Females 50.56 51.31 –0.75 232
Panel C: Field of Study
Finance/Econ 36.74 46.48 –9.74∗∗∗ 240
Marketing/Mgmt 44.36 53.98 –9.62∗∗ 84
Panel D: Financial Literacy
FinLit ≥ 4 36.19 44.63 –8.43∗∗ 220
FinLit<4 47.42 52.33 –4.92∗ 116
Panel E: Type of Fund
% invested all rounds
All fundsall 45.20 47.23 –2.04∗∗ 1,936
Indexall 41.43 48.85 –7.42∗∗∗ 484
Growth/Inc.all 51.87 55.33 –3.46∗∗ 484
Aggr. Growthall 38.85 38.63 0.22 484
Regionalall 48.77 45.63 3.14 484
% invested first round
All funds1st 45.71 50.15 –4.43∗∗ 484
Index1st 34.34 42.85 –8.51∗∗ 121
Growth/Inc.1st 56.17 61.29 –5.12∗ 121
Aggr. Growth1st 42.77 46.29 –3.52 121
Regional1st 46.30 49.97 –3.66 121
Notes: This table shows the fraction of money invested in the female-managed (Column (1)) and male-managed (Column
(2)) fund in our experiment. The difference between the amounts invested in the female- and male-managed fund is
displayed in Column (3). The number of observations is provided in Column (4). Panel A presents results for all subjects
in our experiment, while Panel B contains results for female and male subjects separately. In Panel C, we form subsamples
of subjects by field of study. In Panel D, we divide subjects based on their financial literacy. Financial literacy is computed
based on the number of correct answers in a standard financial literacy test containing six questions on financial issues
(see Appendix D). Panel E displays results for different types of funds and for the first round of the experiment separately.∗∗∗ 1% significance, ∗∗ 5% significance, ∗ 10% significance.
49
Table 8: Items Used in the IAT
Panel A: Gender Items
Female Male
MOTHER FATHER
DAUGHTER SON
GIRL BOY
AUNT UNCLE
GRANDMA GRANDPA
SISTER BROTHER
Panel B: Field Items
Finance Marketing
STOCKS ADVERTISEMENT
DERIVATIVE PRODUCT PLACEMENT
MUTUAL FUNDS MERCHANDISING
STOCK EXCHANGE SALES PROMOTION
CORPORATE BOND BRANDING
MORTGAGE CUSTOMER RELATIONSHIP
INTEREST RATE LOGO
INVESTMENT CONSUMER BEHAVIOR
Notes: This table shows the list of items used in the IAT test. Panel A contains all items used in the gender categories
(female/male). Panel B contains all items used in the field categories (finance/marketing).
50
Table 9: Implicit Prejudice Measures
Measure Mean t-stat 95% Confidence sign. < 0 < 0 > 0 sign. > 0
Interval
(1) (2) (3) (4) (5) (6) (7)
Panel A: All Subjects
d(R) 160.96 10.08 [129.28;192.64] 4 (4%) 8 (8%) 26 (26%) 62 (62%)
d(log(R)) 0.1724 10.95 [0.1411;0.2036] 4 (4%) 8 (8%) 25 (25%) 63 (63%)
d(S) 0.1610 10.08 [0.1293;0.1926] 4 (4%) 10 (10%) 25 (25%) 61 (61%)
Panel B: Compatible Configuration First
d(R) 160.16 6.68 [111.88;208.45] 4 (8.51%) 2 (4.26%) 11 (23.40%) 30 (63.83%)
d(log(R)) 0.1700 7.06 [0.1234;0.2218] 4 (8.51%) 2 (4.26%) 10 (21.28%) 31 (65.96%)
d(S) 0.1602 6.68 [0.1119;0.2084] 4 (8.51%) 2 (4.26%) 11 (23.40%) 30 (63.83%)
Panel C: Incompatible Configuration First
d(R) 161.67 7.50 [118.43;204.90] 0 (0.00%) 6 (11.32%) 16 (30.19%) 31 (58.49%)
d(log(R)) 0.1721 8.39 [0.1310;0.2133] 0 (0.00%) 6 (11.32%) 15 (28.30%) 32 (60.38%)
d(S) 0.1617 7.50 [0.1184;0.2049] 0 (0.00%) 8 (15.09%) 14 (26.42%) 31 (58.49%)
Notes: This table displays differences in reaction times from the implicit association test (IAT). Panel A contains results
for all subjects in our experiment. Panel B contains results for the group that played the compatible configuration first.
Panel C contains results for the group that played the incompatible configuration first. Implicit prejudice measures are
denoted by d(R), d(log(R)), and d(S), respectively. d(R) denotes the difference in the average reaction times R between
the incompatible and the compatible configuration in milliseconds. d(log(R)) denotes the difference in the log-transformed
reaction times R between the incompatible and the compatible configuration. d(S) is computed as the difference in the
speed variable defined as S = 1,000R
between the compatible and the incompatible configuration. Columns (2) and (3)
present t-statistics and the 95% confidence intervals of the average d-measures aggregated at the subject level. Columns (4)
to (7) contain the number and percentage of subjects for which the average reaction time in the incompatible configuration
is significantly smaller (sign. < 0), smaller (< 0), larger (> 0), and significantly larger (sign. > 0), respectively, than in
the compatible configuration on the individual subject level.
51
Table 10: Impact of Subject Characteristics and Experimental Parameters on Implicit Prejudice
Measure Subject Characteristic Obs Mean Std Min Max t-stat p
& Design Parameters d(R)
Panel A: Gender
d(R) Female Subjects 49 158.22 167.79 -203.30 619.93 6.60 0.0000
d(R) Male Subjects 51 163.59 153.07 -107.85 661.38 7.63 0.0000
Panel B: Female and Male Finance Students
d(R) Female Finance Students 18 118.43 180.58 -203.30 438.85 2.78 0.0128
d(R) Male Finance Students 25 223.59 154.66 -66.80 661.38 7.23 0.0000
Panel C: Field of Study
d(R) Finance 43 179.57 172.11 -203.30 661.38 6.84 0.0000
d(R) Marketing 10 224.61 139.06 -15.08 485.90 5.11 0.0006
Panel D: Financial Literacy
d(R) High Literacy 43 176.73 172.30 -203.30 661.38 6.73 0.0000
d(R) Low Literacy 57 149.06 149.88 -107.85 485.90 7.51 0.0000
Panel E: Instructor Sex
d(R) Female Instructor 53 169.03 148.76 -120.80 619.93 8.27 0.0000
d(R) Male Instructor 47 151.85 172.30 -203.30 661.38 6.04 0.0000
Panel F: Time of Day
d(R) Morning Session 35 170.65 148.81 -107.85 619.93 6.78 0.0000
d(R) Afternoon Session 65 155.74 166.10 -203.30 661.38 7.56 0.0000
Panel G: Crowdedness
d(R) Large Sessions 37 170.24 187.51 -203.30 661.38 5.52 0.0000
d(R) Small Sessions 63 155.51 142.15 -120.80 485.90 8.68 0.0000
Notes: This table displays differences in reaction times from the implicit association test (IAT) for different subsamples.
Panel A contains results for subsamples of female and male subjects in our experiment. Panel B contains results for
subsamples of female and male finance students, respectively. In Panel C, we split up our sample by field of study. Panel
D contains results for subjects with high and low financial literacy. Panel E contains results depending on whether the
instructor in the experiment was female or male. Panel F contains results for experimental sessions that took place in the
morning or afternoon, respectively. In Panel G, we split up our sample by number of subjects in each session. d(R) denotes
the difference in the average reaction times R between the incompatible and the compatible configuration in milliseconds.
52
Table 11: Investment Decisions Depending on IAT Result
Panel A: Percentage invested in fund A - Univariate evidence
Female Male Diff. (F-M) t-stat Obs.
Manager Manager
(1) (2) (3) (4) (5)
d(R) > 0 41.51 49.58 –8.06 –3.09 428
d(R) < 0 49.04 43.90 5.13 0.77 56
d(log(R)) > 0 41.52 49.56 –8.04 –3.14 436
d(log(R)) < 0 49.04 42.29 6.75 0.88 48
d(S) > 0 41.52 49.59 –8.07 –3.09 428
d(S) < 0 49.04 43.91 5.14 0.77 56
Panel B: Percentage invested in fund A - Multivariate evidence
(1) (2) (3) (4) (5) (6)
FemaleA –9.464 4.370 –15.894 –8.743 –7.454 –10.388
(–3.15) (0.64) (–3.75) (–2.33) (–1.81) (–2.59)
FemaleA · Prejj –17.283
(–2.27)
FemaleA · SubjGenj 13.376
(2.15)
FemaleA · FinEconj –1.862
(–0.32)
FemaleA ·HighFinLitj –4.492
(–0.71)
FemaleA · InstrGenj 2.121
(0.35)
Controls yes yes yes yes yes yes
Pseudo R2 0.018 0.019 0.019 0.018 0.018 0.018
Observations 484 484 484 484 484 484
Notes: Panel A of this table shows the amount invested in female- and male-managed funds in the investment task
depending on whether subjects exhibit (or do not exhibit) prejudice against females in finance in an implicit association
test (IAT). If d(R) > 0, d(log(R)) > 0, and d(S) > 0, respectively, a subject is prejudiced against females in finance,
and vice versa. Panel B of this table shows results from a censored tobit regression with session fixed effects, where the
fraction of money invested by subject j into index fund A, investmentA,j is the dependent variable. FemaleA is a dummy
variable that takes on the value one, if fund A is managed by a female fund manager, and zero otherwise. All other control
variables are described in Appendix B. t-statistics are in parentheses.
53
Table 12: Spillover Effects of Female Managers
Any Number of Both
Female Females Variables
(1) (2) (3)
FemInCompanyc,t 0.076 0.066
(3.22) (2.52)
NumberFemalesc,t 0.006 0.002
(2.35) (0.69)
Controls yes yes yes
Adj. R2 0.093 0.092 0.092
Observations 11002 11002 11002
Notes: In this table, we use the same baseline specification as in Column (3) of Table 2. In Column (1), we replace our
female dummy, Femalei,t, with a variable that is equal to one if there is any female fund manager at fund company c in
year t, FemInCompanyc,t, and zero otherwise. In Column (2), we replace our female dummy, Femalei,t, with a variable
that is equal to the number of female fund managers at fund company c in year t, NumberFemalesc,t. In Column (3),
we include both variables, FemInCompanyc,t, and NumberFemalesc,t at the same time. The regressions are based on
male-managed funds only. They are estimated with time and segment fixed effects. Standard errors are clustered at the
fund level.
54
Figure 1: Distribution of Funds by Manager Gender
Notes: This figure displays the total number of female- and male-managed funds (bars) and the fraction of female-managedfunds (line). The sample consists of all female and male fund managers responsible for at least one single-managed equityfund from January 1992 to December 2009. Data is taken from the CRSP Survivor Bias Free Mutual Fund Database.
55
Figure 2: Salience of Fund Managers’ Gender
Notes: This figure displays four screenshots for fund information on the Fidelity Magellan fund from the websites of theWall Street Journal, Morningstar, Google Finance, and Yahoo Finance. The manager name is always shown on the firstpage that appears and is circled in red in the pictures.
56
Figure 3: Investment Task
Panel A: Group X
Panel B: Group Y
Notes: This figure displays the information about each fund provided to group X (Panel A) and group Y (Panel B),respectively. Identical information is provided to both groups except for the gender of the fund manager (indicated bythe first name) which is switched between fund A and fund B.
57
Figure 4: IAT Screen
Panel A: Compatible Condition Panel B: Incompatible Condition
Notes: This figure displays the compatible configuration of the IAT (Panel A) and the incompatible configuration (Panel B),respectively.
Figure 5: Reaction Times in the Implicit Association Tests (IAT)
Panel A: Compatible First Panel B: Incompatible First
050
01,
000
1,50
02,
000
RT
1. compatible condition 2. incompatible conditionexcludes outside values
050
01,
000
1,50
02,
000
RT
1. incompatible condition 2. compatible conditionexcludes outside values
Notes: This figure shows boxplots for the reaction times, RT, in milliseconds (ms) for the group playing the compatible configurationfirst (Panel A) and the group playing the incompatible configuration first (Panel B). The vertical line in the box indicates the medianlevel, and the upper and lower hinge represent the 75th and 25th percentile, respectively. The length of the whiskers is determinedby the adjacent value which is still just inside a limit determined by 1.5 times the interquartile range. Extremely low (high) reactiontimes of below 300 ms (above 3 seconds) are set equal to 300 ms (3 seconds).
58